## Empirical Evidence for the Austrian Business Cycle Theory

I have regrouped the series of articles on the ABCT. The present post is subjected to modification and improvement over time, with my readings on the empirical works on the ABCT. Because this article is already too long, the studies not covered here will be treated in another article.

(Last update : December 2014)

1. Short introduction to time series regression.

Most macroeconomic time series are non-stationary or have means, variances and covariances that change over time. They should not be used because non-stationary data are unpredictable and cannot be modeled or forecasted. Time series regression may be spurious since they may indicate a relationship between two variables where one does not exist. Non-stationary behaviors can be trends, cycles, random walks or combinations of the three. There exists several forms of non-stationary processes : (1) Pure Random Walk denoted Yt=Yt-1t, (2) Random Walk with Drift denoted Yt=α+Yt-1t, (3) Deterministic Trend Yt=α+βtt, (4) Random Walk with Drift and Deterministic Trend Yt=α+Yt-1tt. The random walk (with or without drift) must be transformed into a stationary process by differencing (subtracting Yt-1 from Yt, i.e., Yt minus Yt-1) correspondingly to Yt-Yt-1t or Yt-Yt-1=α+εt and then the process becomes difference-stationary. If the data has deterministic trend, detrending is needed. In the case of a random walk with a drift (a slow steady change) and deterministic trend, detrending can remove the deterministic trend and the drift, but the variance will continue to go to infinity. As a result, differencing must also be applied to remove the stochastic trend. The disadvantage of differencing is that the process loses one observation each time the difference is taken. If a series must be differenced once (or twice) before it becomes stationary, then it is said to be integrated of order one, I(1), or two, I(2), and it must have one (or two) unit root(s) (random walk). A stationary series without a trend is said to be integrated of order 0. For detrending we use Yt=α+βtt is transformed into a stationary process by subtracting the trend βt, i.e., Ytt=α+εt. No observation is lost when detrending is used to transform a non-stationary process to a stationary one. When the level of the series is not stable in time, i.e., increasing or decreasing trends, we say that the series is not stable in the mean. If it was variability or autocorrelation, we say the series are not stationary in the variance or autocovariance. If the distribution of the variable at each point in time varied over time, we say that the series is not stationary in distribution. On the other hand, the stationary process reverts around a constant long-term mean and has a constant variance independent of time.

For these reasons, these artifacts should be taken care of by way of unit root tests, such as Dickey-Fuller tests. But auto-correlation (error term at time t depending on previous error term at time t-1) may pose problems. The Augmented Dickey-Fuller (parametric) and Phillips-Perron (non-parametric) are the recommended methods in that situation. ADF and PP regressions attempt to control for serial correlations by including the lagged values of the differenced variable (also called lagged difference terms). In ADF, the number of lags is to be specified by the researcher. To determine the appropriate number of lags, model fit indices can be used, e.g., Akaike’s information criterion (AIC). The null hypothesis is that the variable is not stationary (have unit roots). It is rejected by examine the p-value, which may be problematic to the extent that it becomes smaller the larger the number of observations. When a given variable is not stationary, it must be transformed and used in unit root tests. If the t-test is significant, it is ready to be used in the fixed effect regression.

It is also possible, given stationarity, to use a Granger causality (multiple regression) test. A variable (X) is said to Granger cause another variable (Y) if Y can be better predicted by the lagged values of both X and Y than by the lagged values of Y alone. In other words, the test evaluates whether or not the lagged values of one variable improve the forecasts of another variable. But it would lead to incorrect inferences about causality when there is an error correction process. If variable have unit roots but are cointegrated, then there exists a Vector Error Correction Mechanism. If the series are I(1), we must perform regressions in differences, but this loses one observation. This problem can also be overcome through Error Correction Model (ECM).

Within ECM framework, one key concept is the cointegration. Two processes are said to be cointegrated if there is a linear combination (sum) of the processes that is stationary, which in this case will be called cointegrating relationship. In general, it is not possible to make a meaningful linear regression of one integrated process, e.g., variable I(1), over another. However, regression is possible if the two processes are cointegrated. For example, stock prices may be in random walk but there are portfolios that are stationary (if the aggregate variables Y and X are proportional in the long run, then Zt=Yt/Xt would be stationary). Such cointegrated processes are thus characterized by short-term dynamics (shocks) and a long-run equilibrium : this is the key point of ECM. In consequence of X and Y being both stationary and cointegrated, the residuals must be stationary, and should be tested with unit root analyses before its inclusion in a regression. The residuals Z is obtained by regressing Yt on Xt. This removes the influence that X has on Y. This equilibrium error, Z, captures the error correction by capturing the degree to which Y and X are out of equilibrium. ECM allows inferences based on both levels and first-differences in variables to be possible.

2. Empirical studies.

1. Empirical Evidence on the Austrian Business Cycle Theory
(James P. Keeler, 2001)

Keeler (2001) examines 8 U.S. (post-war) business cycles during the 20th century with standardized quarterly data. His general exposition reads :

According to Austrian business cycle theory, a cycle caused by a monetary shock should exhibit the following patterns: 1) the liquidity effect lowers market interest rates below the natural interest rate, and creates a steeper yield curve at a lower position; 2) investment flows and capacity utilization are systematically increased for more capitalistic production processes in the expansion; 3) short-term interest rates adjust to long-term interest rates with a mechanism related to the cycle; and 4) the expansion phase entails the contraction phase as resource allocations are reversed.

Figure 1 below depicts the ratio RGDP/NRGDP, i.e., real GDP to natural real GDP. The natural real GDP is defined as the real GDP which would have been observed if it had increased at its long term rate of growth. This is supposed to approximate the changes in equilibrium because money over-expansion should produce the impulse that generates the business cycle through actual GDP. Keeler justifies it as follows :

The natural real GDP, measured on a constant growth path between business cycle peaks, is an ex-post estimate of natural real GDP, rather than an estimate based on resource quantity and productivity. This concept permits the possible non-neutrality of money in the long run following a monetary shock, and does not require real values to return to pre-shock values. The income ratio, as a representation of the aggregate cyclical activity, incorporates the change in the equilibrium and serves as a more appropriate measure of an Austrian concept of aggregate economic activity.

Anyway, we see an increase in this ratio until they reach a peak at the 25th quarter before diminishing again after this.

Keeler approximates the market and natural interest rates with short- and long-term interest rates by way of yield curves. The reason given for these variables as proxies reads as follows :

The time period of long-term interest rates, ten years for corporate bonds and U.S. Treasury bond rates, corresponds to the longer time period for the life of capital goods. Assuming that changes in the marginal physical productivity of the existing capital stock are slow, of low volatility and long lived, that behavior is matched by market long-term interest rates. The misperceptions which cause business cycle phenomena, as discussed earlier, include errors in financing capital investments. One of the decisions that investors must make is matching the availability of funds with the cost requirements in the building process for a capital investment project. The risk of that choice may be reduced by aligning the term of borrowing with the occurrence of expected costs of the project during the time-to-build. Long-term financing for capital projects avoids the problems of increases in short-term interest rates before the completion of the project, and appropriate matching of flows of costs and benefits will immunize the project from interest rate risk. The demand for credit for financing capital projects with a long-term flow of benefits and with a long time-to-build will occur in long-term credit markets and be coordinated with the natural rate of interest or expected profit rate.

Graphically, a yield curve is depicted by plotting interest rates in the Y axis (vertical) and maturity length in X axis (horizontal). The three types of yield curves must be distinguished, however. A positive yield curve reflects the higher yield (i.e., interest rates) of longer maturity bonds compared to shorter term maturity bonds (due to entrepreneurial risks associated with time regarding longer-term maturity contract). It is also called normal yield curve because long-term bonds have generally higher interest rates. Conversely, an inverted (i.e., negative) yield curve is one in which the yield is higher in short-term than in long-term, perhaps a sign of a forthcoming downturn. Finally, a flat yield curve indicates identical interest rates for short-term and long-term debts. In a nutshell, a greater/steeper yield curve slope must be associated with greater discrepancy between short and long-term interest rates (i.e., yields). According to Keeler :

… The yield curve arrays rates of return by maturity of the financial instrument, and in general equilibrium, the yield curve would be characterized by a positive slope as term premiums increase with maturity. Then a monetary expansion will have a liquidity effect that lowers short-term interest rates to a greater degree than long-term interest rates (Romer 1996:395–396). Long-term interest rates are affected since they are an average of short-term rates, but the effect is moderated. Relative to its position and shape in equilibrium, the yield curve can be expected to shift down as both short-term and long-term rates fall, and to become steeper as short-term rates fall relative to long-term rates as a result of the monetary shock.

… Consider the short-term interest rate as representing the market rate in a short-term credit market and the long-term interest rate as representing the natural rate for the long-term market. Then the slope of the yield curve is offered in this analysis as a measure to capture Wicksell’s concept of the interest rate differential. In the disequilibrium created by the monetary shock, short-term rates differ from the long-term rates by more than the risk and liquidity premiums justify, and the yield curve is steeper than in general equilibrium. The implied path of these rates is a steep yield curve (a large magnitude for the slope) early in the business cycle and a flatter or inverted yield curve (a smaller magnitude or negative slope) in the recession phase. The yield curve is expected to shift down early in the cycle as both short-term and long-term rates fall, and to shift up in the recession phase as both rates rise.

Figure 2 above illustrates well the overall picture. The slope is positive in the early phase of the expansion but, as time goes on, the positive slope comes closer and closer to zero over the course of the cycle. In some of the historical periods, and between 25-35 in time quarters, the slope is zero or negative. Generally, short-term rate is low relative to the long-term rate in the expansion phase, while the short-term rate rises relative to the long-term rate in late expansion and in recession.

Short-term interest rates fluctuate much more than long-term interests. Keeler cites Bernanke (1990) surveys showing correlations of short- and long-term interests with money supply growth, respectively, 0.29-0.33 and 0.06-0.20. This is important in itself since the long-term interest is a much better approximation of the natural interest rate(s) than is the short-term interest. Bernanke also found that 2 measures of the yield curve, the interest rate spread and the slope of the yield curve, were highly correlated with his measures of monetary policy. In Keeler’s data, slope of yield curve shows the highest correlation with money growth, at 0.55.

… Data for the eight U.S. business cycles show both nominal and real long-term interest rates to be comparatively stable during each cycle. Most cycles show a slight cyclical pattern of a slow and steady rise in the level of the rate through the business cycle peak and then a decline, but three of the eight cycles have long-term rates that are either flat or decrease steadily. Only two cycles exhibit a rise in long-term rates at the end of the cycle. Most cycles show nominal and real long-term rates in a similar range with the exception of the 1980–82 and 1983–91 cycles, which have much higher levels. A variety of inflation adjustments calculates a few negative real long-term rates during the 1950–54, 54–58, 71–75 and 75–80 cycles. The expectations hypothesis implies a pattern of a shift down of the yield curve at the start of the cycle followed by a shift up during the recession, and that is not apparent across the eight U.S. cycles. There is no consistent evidence that long-term rates respond to monetary shocks.

Table 6 is highly relevant for ABCT. Money growth (variable MONEY, which is the compound growth rate of the money supply, using M1 and GDP Deflator chain-type price index) has correlations with changes in slope of yield curve (variable YIELD), i.e., discrepancy between actual market rates and natural rates of interest in the ABCT framework, but only during the 0th through 3rd periods, not after that. This reflects a short but strong liquidity effect. And interestingly, that steepening of the yield curve (variable YIELD) correlates non-trivially with a rise in the ratio of capacity utilization rate in long production processes to short production processes in primary processing industries relative to that in advanced processing industries (variable CAPACITY) only from 3rd through 11th period, but not at the 0th and 1th periods, and only modestly at the 2th period. The same pattern of positively high correlations also exists between YIELD and INCOME (i.e., the variable of the ratio RGDP/NRGDP) with positive relationship only starting at the 3rd period, not before. That seems to suggest that YIELD impacts both CAPACITY and INCOME only after MONEY impacted YIELD. During the 12th through 16th periods, the correlation YIELD-CAPACITY approaches zero, which is not surprising because the market and natural interest rates should come close to each other at the end of the boom and because the ratio long/short production processes is also expected to diminish. Keeler noted, finally : “The cross-correlations do not show significant reversals of correlation within the cycle, but a steeper yield curve is consistently correlated with higher capacity utilization in primary production processes”. Previously, unit root test has been performed.

Each variable was tested for the presence of a unit root. The null hypothesis of the available tests is that the variable has a unit root, and if true the behavior of the variable would be consistent with a random walk process. Then the changes in the value of the variable from one time period to the next would show non-stationarity. The Augmented Dickey-Fuller test was specified with only a constant since all series exhibit no trend but a non-zero mean, and the Phillips-Perron test was specified with a constant and a trend. Tests were performed with four lags. Results are presented in Table 5, including alternative measures of YIELD. The null hypothesis of a unit root is rejected at the 5% level of significance for all variables. The time trend variable is not significant for any of the variables at the 5% level in either the ADF or Phillips-Perron test. Rejecting the null hypothesis of a unit root implies that the variable does not follow a random walk process. The probability of Type II error is large for these tests, but the results, considered with the correlograms, support the notion that the variables are stationary in the forms presented in Table 2, despite the variation in means and variances across cycles.

Table 7 reports the Error Correction Model (ECM) which is a dynamical system with the characteristics that the deviation of the current state from its long-run relationship will be fed into its short-run dynamics. It’s a category of multiple time series models that directly estimate the speed at which a dependent variable Y returns to equilibrium after a change in an independent variable X.

In all specifications in Table 7, SPREAD variable (i.e., long-term rate minus short-term rate) has a large negative coefficient whereas INCOME*SPREAD interaction has large positive coefficient and MONEY*SPREAD is not very consistent. That means the rate of real money supply growth does not have an effect on the rate of adjustment between interest rates whereas the ratio RGDP/NRGDP (i.e., INCOME variable) does have an effect on interest rates adjustement. It’s because the main effect of SPREAD (on short-term interest rate as the dependent variable) was negative, net of the interaction effect, but because the interaction effect has an opposite sign, it means that SPREAD variable is closing due to the effect of INCOME*SPREAD interaction.

Keeler argues nonetheless that the R²=0.40 and R²=0.54 leave a great portion of variance non-explained. Problem is, social scientists aren’t generally aware about the serious shortcomings of R² but well documented by Hunter & Schmidt (2004). According to them, the r (not r²) should be used. Furthermore, the main problem with significance test (t-test, or F-test) is to be sensitive with sample size (being significant if N is large). But overall, the tenets of ABCT seem not to be rejected here.

2. Relative Prices and the Business Cycle
(James P. Keeler, 2001)

Keeler (2001) examines U.S. data for the time period 1959:1 through 2001:2. Federal funds interest rates existed at the beginning of this period already, but the federal funds market, with its role in monetary policy, starts its development at 1966. Several variables are the targets of analysis. The slope of yield curve variable is given by the ratio of long-term rate to short-term rate; a positive yield curve reflects the higher yield (i.e., interest rates) of longer maturity bonds compared to shorter term maturity bonds (due to entrepreneurial risks associated with time regarding longer-term maturity contract). The income variable is given by the ratio of real GDP to natural real GDP; the natural real GDP is defined as the real GDP which would have been observed if it had increased at its long term rate of growth. The monetary policy variable is a mix of borrowed and non-borrowed reserves in total reserves (NBRX) suggested by Strongin (1995) and has been considered as providing a good measure of the effects of monetary policy. Two measures of resource allocations are provided. NRSRS is the ratio of quarterly investment in Non-Residential Structures to investment in Residential Structures and IETE is the ratio of investment in Industrial Equipment to investment in Transportation Equipment.

Keeler submitted all variables to a unit root test, the Augmented Dickey-Fuller. The null hypothesis of unit root was rejected for income, slope of yield curve, and resource allocation, but not the Strongin’s measure of monetary policy. Most of the series exhibited autocorrelations that quickly decrease to zero. The results from Vector Autoregressive (VAR) modeling reads as follows :

The VAR method is a “backward-looking autoregressive structure” (Bagliano and Favero, 1998), and still subject to the Lucas Critique; that the estimated responses within the system are specific to the policy regime. Over the period of this sample, there certainly were developments and changes in monetary policy, from reliance on lending at the discount rate, to a free reserves target without specific goals, to interest rate or monetary aggregate targeting with specific goals (Strongin, 1995). … An advantage of the VAR is that the restrictions that must be imposed for identification of equations involve fewer assumptions about the equation structure. A Choleski factorization for example, orders the variables in the system’s causal sequence, but does not add further constraints on the coefficients. The flexibility of the VAR method permits modeling a changeable process, as well as structure in terms of the choice of variables and identification. …

Each of the four equations in the VAR estimate of the business cycle model had a relatively high R-squared value. For each equation, the lags of the dependent variable are significant. Lagged income is significant in the NBRX equation which suggests the endogenous nature of the money measure. Lagged income is significant for only one coefficient in the SLOPE equation, showing some evidence of the endogenous nature of the interest rate adjustment mechanism. One coefficient of INCOME is also significant in the RESOURCE equation, implying that the phase of the business cycle affects the resource allocation between primary and advanced production processes. One SLOPE coefficient is significant at the 10% level in the INCOME equation, providing only slight evidence that relative prices and resource use guide the cyclical behavior of aggregate income. The residual correlation matrix shows especially low correlations between the NBRX residual and other equation residuals, and generally low correlations.

Impulse Response Functions were derived from VAR models. The IRFs display (in Figure 6) the dynamic responses of the endogenous variables of the model to a one-standard-deviation shock to the error terms of system equations (here, the NBRX). At the beginning of the cycle, the yield curve becomes steeper but during the expansion phase of income increase, the yield curve flattens and returns toward its initial value. The investment in non-residential structures rises relative to investment in residential structures after the shock in NBRX, and then declines as the cycle progresses. The response of income ratio variable to NBRX innovation is also consistent with the ABCT prediction, as the actual real GDP rises relative to natural real GDP, then falls toward the original ratio.

Figure 7 shows the variance decomposition for each variables. The decomposition shows the contribution of each variable (in different equations) through the cycle. For example, in the INCOME equation (upper left) there is an increasing role of the variable slope, which peaks at about 30% after 10 periods. The variable INCOME shows a response in the NBRX and SLOPE equations.

3. Austrian Business Cycle Theory : Evidence from Scandinavia
(Rasmus Anker, 2011)

Anker (2011) attempts to empirically test the ABCT in Denmark, Norway and Sweden with the economic data collected ranging from first fiscal quarter of 1980 to fourth fiscal quarter of 2010. The study uses the following variables : REV for the ratio RGDP/NRGDP, DEP for the ratio consumption/investment expenditure, PRIX_REL for the ratio consumption/production price index, SPREAD for the long-term rate minus short-term rate. Concerning DEP, it’s calculated by consumption aggregates as a percentage of total economic activity (consumption) and the investment percentage calculated by 1 minus consumption. The ratio is then calculated as the consumption divided by investment.

Figure 4.1 presents the trend RGDP/NRGDP ratio :

In the expansionary phase real GDP rise relative to the natural level and the ratio and moves above 1, whereas in the recession phase the real GDP decreases relative to its natural level and the ratio moves below 1.

Figure 4.3 depicts the trend consumption/investment expenditure ratio :

Because the ABCT does not expect the consumption to rise at the very beginning of the boom, the ratio will begin to peak later during the boom phase, probably extending to the recession phase. This illustrates the relative increase in capital goods production which is followed then by a relative increase in consumption goods production.

Figure 4.4 shows the ratio consumption/production price index. It could have been expected that the price of capital goods would go up relative to the prices of final goods due to a change in relative demand for more capital goods. At least, if everything else is constant. Its trend, when compared with phases of expansion/recession pictured in Figure 4.1, suggests no relationship between these two variables, as it will be seen in econometric tests displayed further below.

The “term spread” or SPREAD variable is best illustrated as follows :

The remarkable negative spread in figure 4.6 is a consequence of the ERM crisis of 1992 and 1993. More interestingly the figure shows a pattern where shocks increase the spread and then gradually declines and becomes negative, before experiencing another sharp increase. This can be seen most clearly in the period between 2002 and 2010.

Having presented the data, he makes use of regression models in the following configuration :

REVit= β0 + β1DEPit + β2PRIX_RELit + β3SPREADitαi + uit

where REV is the ratio RGDP/NRGDP, DEP the ratio consumption/investment expenditure, PRIX_REL the ratio consumption/production price index, SPREAD the long-term rate minus short-term rate, β0 the constant and the following β (or Beta) are the coefficients of independent variables, αi is the deterministic constant (i.e., intercept) specific to each of the three countries, uit being the error term, i is the number of countries (=3) and t the number of quarters from 1rst quarter 1980 to last quarter 2010 (=124) so that 3*124=372.

But a unit root test is first performed to test the null hypothesis of the variable being non-stationary and having unit root. As Anker makes it clear : “Stationarity is a basic requirement for time series modeling which implies that the mean, variance and autocorrelation structure of the variables are independent of time”. The result is that the null (H0) is rejected due to the significant p-values for all variables except changes in ratio consumption/production price, denoted ΔPRIX_REL.

ADF and PP tests are commonly used to remove any autocorrelation (also known as serial correlation or autoregression) which takes form of correlated error terms, i.e., error term depending on the previous error term. ADF regression includes lags of the first differences of yt. It appears that Rev, Dep and Spread are significant, thus are stationary. But Prix_rel is not, and for this reason should not be used in the fixed effects regression (see Table 4.3). Nonetheless, the variable has been first-order differenced, yielding a new variable (ΔPrix_rel) that appears stationary. Their regression model in Table 4.3 and related discussion can be read :

As for the estimated α-values specific to each country; Denmark has 0.00935, Norway has -0.02889 and Sweden has 0.00432. This indicates that the joint constant in table 4.2 minus the country specific α-values, are lower for Denmark and Sweden and higher for Norway. The intuitive explanation behind this is that the variables have more explanatory power in Denmark, as less information is stored in the constant compared to Norway and Sweden. …

Post-estimation verifies the fixed effects model and the results can be seen in appendix E. The Hausman test for testing the fixed effects model against a random effects model shows a strong significance in favor of the fixed version. Serial correlation test shows that there is no serial correlation in the residuals with the test unsuccessful in rejecting the null hypothesis for just that. Wald tests for testing the linear relationship of the explanatory variables yields significant results for the consumption to investment ratio and the term spread but not for the relative price ratio as expected. Finally, testing for the country specific effects, results in a rejection of the null hypothesis that these are jointly significant.

According to the theory, the influence of Dep and Spread, should be negative. This is exactly what we see in the fixed effects model. An increase in the ratio consumption/investment expenditure (Dep) has a negative impact on the ratio of actual real GDP to natural real GDP. In recession, natural GDP outgrows actual GDP because of the liquidation of malinvestments. This is accompanied by an increase in the term spread, and this is why it also has a negative coefficient on the ratio actual/natural real GDP. When the term spread increases it brings real economic activity to accelerate compared to its natural GDP until the short term market interest rate converges to its natural level.

Globally, Anker and Bismans & Mougeot present the same feature. But one would like to question why the coefficient (probably unstandardized) of Dep and Spread are all so small, and yet statistically significant, either looking at t-test or at the p-values, considering they are in fact highly dependent on sample sizes, especially with total observation of NT=283. At the very least, the R² of 0.153 suggests the model having good explanatory power if expressed in r by square rooting the R².

4. Austrian business cycle theory: Empirical evidence
(Francis Bismans & Christelle Mougeot, 2009)

Bismans & Mougeot (2009) examined four countries, France, Germany, Great Britain, and USA between 1980 and 2006. Data come from Eurostat and OECD. The method is similar to that subsequently used by Anker (2011) with exactly the same variables’ names.

The ratio RGDP/NRGDP is expected to be 1 at equilibrium, so that any deviation from 1 corresponds to changes in economic activity through the cycle. Figure 1 depicts the following :

Figure 3 shows the ratio consumption/production price index. Like in Anker, there is no apparent relationship between the changes in structure of relative prices and economic fluctuations or production structure changes. But Figure 2 shows the consumption/investment expenditure ratio, used again as an indicator of the production structure distortion trough the cycle, and the pattern seems coherent with the ABCT :

Figure 2 shows that in each country, this ratio tends to increase during the last steps of expansions and to lower at the end of recessions. This pattern seems to be true to the Austrian hypothesis according to which the beginning of expansion is characterised by a relative increase of capital goods production, whilst the rise of consumption goods production accelerates later. Figure 2 also shows that the maxima of the ratio of consumption expenditures to investment expenditures are often reached during the quarters of recession or during the quarters just after recessions. This observation corroborates the Austrian hypothesis of overinvestment liquidation marking crisis.

Since long-term rates are equal to the weighted average of short-term rates plus a risk premium, credit expansion lowers short-term rates higher than long-term rates. So, to repeat, the term spread (equivalent to SPREAD variable in Keeler’s 2001 study) should rise at the beginning of the boom, gradually decreasing, and perhaps becoming negative during the quarters just before a recession. Bismans & Mougeot noted :

In other words, the term spread inversions mark the turning points of the aggregate economic activity. When the term spread decreases, the structure of production becomes less roundabout as entrepreneurs reallocate resources away from production goods to consumption goods.

At the beginning of expansions, the short-term interest rate is smaller than the long-term interest rate (the difference is one of the order of two to four points). Then, the difference filled itself and is reversed before recessions.

Like what has been observed in Keeler as well, the authors argue : “Temporary credit expansions involve decreases in short-term interest rate that slightly influence long-term interest rate. There is no consistent evidence that long-term respond to monetary shocks”. Their econometric analysis, similar as that of Anker, reads :

Unlike L-L-C, I-P-S, or M-W, the Hadri’s null hypothesis is that the variable’s process is stationary, i.e., the Hadri must have non-significant p-value whereas the others must have significant p-value. This condition holds for REV and SPREAD. But while the Hadri is unfortunately significant for DEP, all other tests displayed evidence of stationarity for DEP. PRIX_REL, in the last panel, however shows no stationarity in all of the 4 different tests under column “Niv.” (levels of the series) but it shows evidence of stationarity in all tests under column “Diff.” (i.e., first-differenced or lagged series) which means PRIX_REL is not stationary in levels of its values whereas ΔPRIX_REL is stationary.

The constant β0 (Eq. 4) is positive and equal to 1.068. In comparison, the estimated values of the fixed effects (αi parameters) specific to each country are 0.00079 for France, −0.01926 for Germany, 0.0314 for Great Britain and −0.01313 for the USA. Consequently, the joint constant, that is β0i, i=1, …, 4, is lower for Germany and the USA than for France and Great Britain. This difference implies that the explanatory power of the variables is greater in Germany and in the USA. …

The explanatory power of term spread and relative expenditures differs among countries. The joint constant of the present model is lower in Germany and in the USA than in France and Great Britain. This difference means that the explicative variables are more decisive in Germany and the USA than in the other countries. This result confirms Bernanke’s (1990) conclusion that the explanatory power of the term spread is higher in Germany and the USA than in the other countries (France, Japan and Great Britain). He explains this difference by the difference of monetary market regulation.

Again, problems with significance testing is related with sample sizes and/or data points. But if that can be taken at face value, then, the impact is really non-zero. More informative would be to obtain the magnitude of the effect size, which is not expected to be impacted by sample size artifacts. In any case, they conclude :

The hypotheses concerning the path of the term spread and of the relative expenditures are empirically confirmed. According to the Austrian business cycle theory, their influence is expected to be negative. This result is obtained empirically with the negative sign of the coefficients. … In other words, the growth of the difference between the long-term and the short-term interest rates tends to lower the difference between actual real GDP and natural real GDP. This lowering reflects an acceleration of the economic activity to meet its natural level. This result confirms the main Austrian hypothesis that expansion is created by a decrease of the interest rate under its natural level and lasts until the market interest rate moves towards the natural interest rate.

Here again, Bismans & Mougeot (2009) confirmed the absence of relationship between consumption/production prices and business cycles and agreed with Keeler (2001) on the notion that prices will not convey the signals of real behaviour. They pointed out that Mises (1949) was right in abandoning his first idea (in 1912) of such relationship. Mises (1966, p. 561) insists on the importance, not of the high/low price levels, but of the difference between costs of production and anticipated prices of the products, i.e., profits. This crucial point has not been well understood by Wainhouse (1984), Anker (2011) and Lester & Wolff (2013), however. The simple fact that the ratio consumption/investment expenditure followed a path readily predictable by the ABCT argues in favor of the ABCT. It’s just that the ratio consumption/production price is probably affected by other external factors and thus does not faithfully reflect the magnitude or pattern of distortion of the ratio consumption/investment expenditure or the interest rates gap. Another possibility is  the applied methodogy (e.g., variables used). For instance, Young (2012) found evidence that the more roundabout industries have experienced a steeper increase in their prices during the year 2000s on the onset of the subprime episode.

5. Empirical Testing of the Austrian Business Cycle Theory: Modelling of the Short-run Intertemporal Resource Allocation
(Tobias Helmersson & Karl Selleby, 2009)

Tobias & Karl (2009) modeled the ABCT by evaluating the impact of interest rate gap, money supply and credit had on the intertemporal resource allocation, denoted C/I, or the ratio of consumption to investment, given a short-run framework, by using quarterly data. The data in use is collected from the Bank of England and UK National Statistics (all seasonally adjusted) beginning at first quarter (Q1) 1984 to first squarter (Q1) 2006 with 421 observations in total. The business cycle phases studied in their models are splitted based on unemployment levels.

In their regression models (equations given at page 17), the interest rate gap, credit, ΔM0, ΔM4 are the independent (predictor) variables with ΔC/I being the dependent variable. M4 is expressed here as M4 aggregate minus M0. The beta (β) without the letter D indicates expansionary phase whereas the beta with letter D indicates the recession phase.

They followed Carilli & Dempster (2008) recommendation of using consumption/savings ratio as proxying the natural interest rates. The natural interest rate is thus calculated as the average propensity to save divided by the average propensity to consume. This allows the intertemporal relationship in purchasing patterns and the opportunity cost of holding money to be fully incorporated in the model. The Consumption index is “Household final consumption expenditure” and the Investment index is “Gross fixed capital formation: Business Investment”. As for M0 and M4 :

M0 is the narrow monetary aggregate of the UK, and M4 is the broad money supply measure. M0 cover notes and coins in circulation, banks’ vault cash holdings and bankers’ deposit at Bank of England (BOE), also known as BOE’s and the government’s liabilities (Janssen & Andrews, 2005). M4 contain private sector holdings of sterling pounds, sterling pounds in deposit (here equivalent to certificates of deposit and debt securities up to five years) at banks and financial institutions, and sterling shares issued by financial institutions (Burgess & Janssen, 2007). M0 consists of roughly five percent of the amount of M4 (Congdom, 2007, p. 319).

Unlike Anders (2011) and Bismans & Mougeot (2009) these authors haven’t conducted a test for unit-root, perhaps because they estimate the correlation between changes in the dependent var. with changes in each of the independent var., and they expect it to remove problems of non-stationarity and autocorrelation. The expectations of the models are presented.

The variable M0 is expected to have a positive impact on C/I, as the narrow money supply is very liquid and flexible but also because M0 is less useful for business projects that need long term financial capital (e.g., credit to finance structural investments). The positive effect on C/I should be less in recession because of employment uncertainty and debts. M4 should have negative effect on C/I ratio. The reason is because this monetary aggregate is a less liquid measure of money, consists of time deposits in banks and financial institutions, making it more suitable for long-term projects and investments. In recession, M4 is supposed to have negative impact on C/I as well, but less in magnitude. The interest rate gap is assumed to have more influence on investment than consumption, thus lowering C/I ratio in expansion. Although in recession the market rate should increase relative to the natural rate, the governments may attempt to stimulate the economy through money injection. Still, credit is expected to have more (positive) impact on investment during the boom than during the bust. The authors assume that even if the credit get distributed more to consumption, the consumption will not increase much, as output is fixed in the short run, where inventories diminish and prices increase, whereas businesses can start increasing investment into the structure of production by using credit.

The results are displayed below. Significance does not matter because it conflates sample size with effect size. Yet, it still give an information about the relative strength of the independent variables. The variables with larger t-stats or lower p-values must have larger effect. The variables may not be expressed in the same unit-scale (e.g., interest rates compared to money aggregates). The betas are unstandardized. In each model, β0 is the constant (or intercept) for the expansion, and β1D is the same for recession.

Given table 5, ΔM0 in negative in the two phases, but the magnitude of the beta is much larger in recession. According to the authors, M0 should have been positive in expansion and recession (although stronger in expansion) because M0 is closer to consumption and increases in M0 would have impacted C more than I. So, C must increase relative to I, causing beta to be positive. And because we don’t see it during the expansion, that is not consistent with ABCT. As we will see, the beta for ΔM0 is negative only in this model, perhaps different variables are included in the subsequent models. Different variables are controlled (i.e., held constant) while measuring the impact of ΔM0. So, when M4 is taken into account, it looks as if M0 does not predict increases in C/I anymore. Next, ΔM4 is negative in expansion, positive in recession. Because the expectation was a negative sign in both phases, this pattern follows ABCT prediction (only in expansion) because M4 should be closer to I than to C, causing I to increase relative to C in expansion, and thus the ratio C/I must decrease as I increases.

Given table 6, ΔM0 is positive in expansion, negative in recession. According to the authors, this is consistent with ABCT only in expansion. The interest gap variable is the natural rates minus market rates, that is, higher values indicate less interest rate discrepancies. And the interest gap has positive sign in expansion (lower discrepancy decreases I relative to C thus causing C/I to be higher) and negative sign in recession (lower discrepancy decreases C relative to I, causing C/I to be lower). The ABCT predicts this pattern in expansion, not in recession. The interest rate gap variable must have a positive sign in both expansion and recession, according to the authors.

Given table 7, ΔM0 is positive in expansion, negative in recession. According to the authors, this is consistent with ABCT only in expansion. The changes in credit should increases I relative to C, causing C/I to be lower. This is what we see. The sign in β4 is indeed negative. However, in β5 it is positive, which means credits increase consumption more than investment in recession, even though the magnitude of the beta is much smaller. An indication that consumption increases (in recession) had been much smaller than investment increases (in expansion). In all likelihood, the positive sign of β5 can be explained by keynesian stimulus. We can point out a mistake in the authors’ interpretation since they have written that M0 is positive in both phases and credit is negative in both phases, which is not true with regard to their tables.

Despite the (adjusted) R² being under-estimation of the real-world effect size, the (adjusted) R² is much higher in Table 6, i.e., the model with interest rate gap has the best explanatory power among all of the models. The SQRT of 0.228 is 0.477. A meaningful effect size. This was also suggested by the p-value of interest gap variables, all being 0.000.

While the empirics don’t follow ABCT prediction in recession phases, they are coherent with ABCT in expansion phases. Why the data shows this pattern is obvious. It’s due to external forces in the recessions. Similar speculations have been advanced by the authors in their thorough explanation of the conflicting findings :

There is also the possibility that using a short-run model creates problems in finding significance as it fails to capture the underlying predictions observable in the long run. Logically, a theory that is designed to have long run impact due to accumulating changes does not have the identical outcome when the incremental stages are examined. Narrowing the time frame of observation naturally leads to more uncertainty, hence observations in the short run need not necessarily match observations in the long run.

The narrow monetary aggregate M0 was predicted to track fluctuations in consumption better than the investment part of the C/I ratio. While consumption in the short run is rigid due to the scarcity of consumer goods available as explained, it can increase incrementally in relation to investments. The broader money aggregate, M4 was expected to have stronger effect on investment, which the data supports. Rothbard (1978, p. 153) defined the Austrian school’s supply of money concept: “all entities which are redeemable on demand in standard cash at a fixed rate”. Neither of the two considered official monetary aggregates in the UK corresponds to this statement, as M0 do not include redeemable time deposits while M4 includes illiquid deposits with constraints on withdrawal. Perhaps due to not being appropriate “Austrian monetary aggregates” both give conflicting coefficient signs compared to predictions. The finding that M4 and M0 as predicted affect C/I differently is important in further discussions of the ABC theory, with a proper monetary aggregate located in-between M4 and M0. …

The effect of liquidation force can be seen to some extent. The ambiguity found in some of the results when predictions are made for variables in recessionary periods, could be the result of conflicting liquidation and stimulating forces. These are arguably common as policies can distort the fundamental market movements.

And since the ABCT is an accounting of how the expansion occurs and develops itself, these findings are in line with the theory.

6. An Empirical Examination of Austrian Business Cycle Theory
(Robert Mulligan, 2006)

Mulligan (2006) uses the (January)1959-(March)2003 monthly data, reported by the U.S. Department of Commerce Bureau of Economic Analysis, to examine the relationship between real consumable output and the interest rate term spread. The price index was used to obtain real personal consumption expenditures, which is the measure of real consumable output. The term spread used is the ten-year constant maturity Treasury bond rate minus the three-month Treasury bill secondary market rate, from the Fed of St Louis. When the term spread decreases, the structure of production becomes less roundabout as entrepreneurial managers reallocate resources away from producers’ goods toward consumers’ goods. If the term spread is interpreted as a measure of the real interest rate, the cumulative term spread can be interpreted as the real return over time.

Whenever interest rates rise, higher rates of return in production are necessary to compete with financial instruments, such as relatively higher-yielding government bonds. This is manifested in a shifting of resources away from early stages of production to later stages, and can be shown as a shortening of the base of the Hayekian triangle.

From Mulligan’s perspective, monetary over-expansion must cause a decline in real (consumable) output which is “the subjective use value extracted by each consumer” because it cannot be easily reallocated. This is due to capital degree of specificity :

When the interest rate rises, capitalists should liquidate their own productive activities to the extent possible, and lend the money out to take advantage of the higher return. However, physical capital comprises illiquid assets, and once savings is invested in productive activities, it cannot be extracted without delay and loss of value. Physical or installed capital is characterized first by its complementarity with other components of an entrepreneurial production plan, and only secondarily by its substitutability in alternative plans …

As a rule more illustrative than actually descriptive, the need for additional complementary resources for production is approximately proportional to the amount already in use, for example, the amount of physical capital already installed. Thus more capital installed means more additional resources required, so the demand for additional credit accelerates. If the supply of additional credit remains steady as the demand for it increases, the interest rate must rise.

Mulligan also performs the unit-root and cointegration tests (see Murray, 1994; Smith & Harrison, 1995). Those two conditions should be met before ECM application. The variables involved in ECM should be cointegrated, and they are said to be I(1) or I(2) if they contain 1 or 2 unit root(s). Their data points have means, variances and covariances that change over time, that is, the variables’ levels are not time-invariant. If two series are said to be cointegrated, they must be integrated of the same order (e.g., avoiding something like I(0) for variable1 and I(1) for variable2) and a linear combination (e.g., the sum) of them must be stationary (i.e., yt-βxt=ut where β is the cointegrating coefficient). This equilibrium error, ut, which captures the error correction by capturing the degree to which Y and X are out of equilibrium, is sometimes denoted zt and can be obtained by regressing yt on xt in order to remove the influence of x on y, so z (or u) is the portion of Y (in levels) that is not attributable to x. Cointegration is needed to establish long-term (equilibrium) relationship.

Because stationarity of money growth has been confirmed, the next step can proceed. He applies ECM, a multiple regression analysis that attempts to predict the effect of X (independent var.) on Y (dependent var.) controlling for lagged X (i.e., Xtime-1) and lagged Y (i.e., Ytime-1). Thus, the model might look like : Yt=α+β0Yt-11Xt2Xt-1t. The constant α expresses the contemporaneous effect of X on Y whereas the long-term effect of X on Y at t+1 can be obtained by X value minus lagged X value. See Robin Best (2008).

Error-correction models provide estimates of both a structural or equilibrium process toward which adjustment is generally effected, and the error-correction or disequilibrium adjustment process through which adjustment is made toward the hypothesized equilibrium. Even if one rejects the reality of any hypothesized equilibrium, estimates of the disequilibrium adjustment process still warrant interest. The error correction model consists of two parts, a structural equation which defines the long-term equilibrium process, and a short-term disequilibrium adjustment process. The residual of the structural equation is an estimate of the disequilibrium in any given time period.

The Johansen-Juselius tests for cointegration reveals a stable, cointegrated relationship between real consumable output and the cumulative yield spread. The next step can proceed.

The OLS estimate also allows for a test of the hypothesis that a lower interest rate accompanies a permanently lowering of the level of real consumable output, the key assertion of Austrian business cycle theory. This interpretation assumes that interest rates fall only due to expansionary monetary policy and not due to general lowering of time preference. The adjusted R square is 97 percent. The intercept and coefficient on the cumulative term spread are both positive and significant. Coefficient values of 6.862 for the intercept and 0.162 for the slope indicate that a one percent increase in interest rates permanently raises consumption expenditure by 955.3 billion chained 1996 dollars each month the higher interest rate persists.

Perhaps more revealingly, a one-percent decrease in the cumulative term spread, such as results from policy induced monetary expansion, has on average decreased real consumable output over the long run by the same amount. The results of the t-test on the cumulative term spread provide strong empirical confirmation of Austrian business cycle theory. This amount is more than great enough to account for any historic recession. Further, the output measure used here, real consumption expenditures, comprises only approximately 70 percent of GDP, thus any impact on real consumption implies a somewhat greater impact on total real output.

The estimate of the vector error correction model (VECM) is reported in Table 5. To facilitate interpretation, the VECM is normalized with respect to and solved for consumption. Estimated coefficients of the cointegrating equation are similar in sign and magnitude to those found by OLS. The VECM intercept and slope coefficients 7.120 and 0.136, indicating a one-percent decrease in the cumulative term spread decreased real consumable output by approximately 1.2 trillion 1996 dollars for every month the term spread falls. [7] This is significantly greater than the amount indicated by OLS, but the two estimates are reasonably consistent. The t-test on the VECM estimate of the structural equation provides further evidence in support of Austrian business cycle theory’s key assertion that lowering the real interest rate lowers real consumable output over the long run.

[7] The impact of the interest rate on consumption is evaluated by taking the slope coefficient estimate of 0.136, multiplied by 1.0 representing a decrease (increase) of the interest rate by 1 percent for one month, and taking the antilog of 0.136, which equals a 1.15 billion chained 1996 dollar loss (gain) in consumable output (consumption spending) for every month the cumulative term spread is lowered by 1 percent. The longer the interest rate is kept 1 percent below the sustainable market rate, the greater the impact on the cumulative term spread and thus on real output. See table 6.

Table 6 shows that whenever the term spread has been lowered significantly below its average value, real consumable output is permanently lowered by a significant amount.

Graphs of the impulse response functions are presented in Figure 2. The upper-right-hand graph is the one of interest for Austrian business cycle theory. It indicates that over the period studied, a one standard-deviation increase in the term spread has resulted, on average, in an upward adjustment of approximately .004 in the logarithm of consumption, equivalent to 1.004 billion 1996 dollars after eight years or 96 months. A one standard deviation decrease in the yield spread decreased real consumable output by an equivalent amount, on average.

The variance decomposition functions are presented in Figure 3. The upper-right-hand graph indicates that after 96 months, nearly 45-50% of the variance in real consumption (LC) expenditures has been attributable to variation in the cumulative term spread (LR). Conversely, no variation seems to be transmitted from consumption to the interest rate, given the lower-left-hand graph. The author says this is not surprising since interest rates are set by policy, and that output, including consumption, responds unfavorably to policy initiatives.

7. The Austrian Business Cycle: a Vector Error-correction Model with Commercial and Industrial Loans
(Robert Mulligan, 2005)

Mulligan (2005) uses data from the FRED-II series. Output Index is the industrial production, reinitialized at January 1959 = 100. It estimates the value added in mining, manufacturing, and utilities industries, excluding virtually all services. Consumption Index is the annualized real personal consumption expenditure, observed monthly for January1959-March2003. Consumption spending includes both consumption goods and services. Investment Index is imputed based on the difference between total real output and real consumption. Monthly percent growth rates are computed for the industrial production index and the consumption index. It is assumed that any real output produced which is not consumer goods is producer goods. The percent growth rate of the consumption index is subtracted from the percent growth rate of the industrial production index. The resulting difference is taken as the imputed percent growth rate for real investment. Starting with January 1959 = 100, the imputed real investment index for period t+1 is constructed by multiplying the index for period t times one plus the imputed percent growth rate. Credit Index is the commercial and industrial loans at all commercial banks, observed monthly. This nominal value is adjusted for changes in the price level. The producer price index (PPI) for all commodities is used as a deflator, observed monthly. The deflated series is converted to an index with January 1959 = 100. The credit variable is not without problems, as Mulligan notes :

A potential criticism of commercial and industrial loans as a measure of credit expansion is that credit injected through capital markets is ignored. During expansions, credit is injected into the stock market, dramatically inflating equity prices. However, almost all credit allocated to capital markets goes to purchase already-issued equities. Additions to capital-market credit contribute to inflating stock market prices and indices, but very little becomes available for investment projects, through initial public offerings and initial sales of corporate bonds and commercial paper. Furthermore, commercial and industrial loans proxy the two latter forms of credit very well.

The ADF test results with 48 lags indicate output, investment, and commercial and industrial loans are all I(1), but that consumption may be 1(2) or integrated of higher order. This outcome is detrimental for ECM which needs variables to be integrated of the same order. However, Phillips-Perron tests indicate that all variables are I(1). The Johansen-Juselius test shows the cointegration needed for use of ECM. The ECM is employed to estimate the stable long-term relationship between the variables.

Table 4 shows that the effect of credit on industrial production is positive but only significant at 10% level. Mulligan believes this is because the effects cancel out, being positive during the boom and negative during the bust. The positive coefficient (significant at 5% level) of consumption index shows that credit expansion permanently increases real consumption expenditures. This also suggests that consumers save less in response to lower interest rates. The slope coefficient for investment is negative and very significant, and this indicates that credit expansion permanently lowers investment expenditure.

Adjusted R-squares for the disequilibrium adjustment processes in the cointegrating vector are very low. In spite of the low R-squares, disequilibrium adjustment terms [Θ, Ψ, and Ξ] are significant at the 5% level, only in the disequilibrium adjustment process for consumption. This is an especially interesting result, which is easy to account for according to ABC theory. Apparently market disequilibria, measured by non-zero residuals in the three structural equations, effect correction chiefly through changes in consumption spending. Below-equilibrium consumption, measured by positive residuals in the consumption equation, results in positive adjustments to consumption accompanied by decreases in industrial output and investment, as indicated by significantly negative coefficients on the disequilibrium adjustment terms. Consumption itself adjusts upward, as indicated by the significantly positive coefficient. Little or no adjustment occurs through total output or through investment, suggesting that credit-induced increases in consumption generally occur at the expense of investment and output, rather than as additions to them. Consumer behavior is highly responsive to market disequilibria, but producer behavior exhibits much more inertia, likely due to the fixed capital embodied in the production structure.

The Impulse Response Functions (IRFs) show that IP, consumption and investment go down by -40%, -12% and -25% after a positive shock in commercial and industrial loans after 48 months.

8. A Hayekian Analysis of the Term Structure of Production
(Robert Mulligan, 2002)

To the extent that the Error Corection Model evaluates how fast the disequilibrium error (i.e., short-term changes) of a given relationship adjusts itself to its long-term relationship, Mulligan considers it relevant for the study of the ABCT. He uses the data from the U.S. Department of Labor which provides seasonally adjusted monthly sectoral employment data for the 1959–2000 period. Interest rates for several maturities, ranging from three months to five years, were used to capture the apparent implicit time preference latent in the nominal term structure. The purpose is to examine how fast the employment in industrial sectors responds to changes in interest rates. This point is very important because if the evidence was supportive, then, any hypothesis that does not involve the role of interest rates in resource reallocation will be unable to explain the data :

If resource reallocation occurred because of changes in intertemporal consumption preferences or technological change, that reallocation should not be systematically related to interest rates, especially nominal interest rates. In the sample period of over 40 years, dramatic technological advances and changes in consumer time preference certainly occurred, due, for example, to demographic changes in age distribution. Unlike policy-induced credit expansion, these factors do not play any systematic role in driving the business cycle. Technological change is generally interest-neutral (Garrison 2001, pp. 59–62), while credit expansion never is. Also, like technological change, changes in time preference never cause unsustainable reallocation of productive resources.

The technique ECM must fulfill the assumption of cointegration, given non-stationary time series, e.g., I(1). Obviously, the variables must be integrated of the same order (both I(1) and not I(1) and I(0) for example). Here’s a discussion of some of these issues :

The finding that mining may be I(0) indicates it might be deleted from the vector error-correction model, but mining was retained in the model because it remains a part of the nation’s employment statistics, though declining in importance throughout the sample period, and because there is some support for the conclusion that mining is I(1). The null hypothesis of a unit root is always rejected for the first-differenced series, demonstrating all are integrated of order one [I(1)] and not of higher order. Somewhat surprisingly, the interest rates are found to be I(1). A priori, interest rates are expected to be I(0). This would not present any difficulty for interpreting vector error-correction models that include I(0) interest rates and I(1) employment rates, because in order for the I(1) employment rates to enjoy stable, long-run relationships with any I(0) series, the employment rates would have to be cointegrated. Since more than five cointegrating vectors are found, there must be some cointegration among the employment rates.

The five-equation model, where the dependent variables are interest rates, is reported in Table 3. Each cointegrated vector is normalized with respect to, and solved for, different interest rates with maturities ranging from six months to five years. Cointegrated vectors are solved for the interest rate to which they are normalized. The coefficients are interpreted as inverse elasticities of employment with respect to interest.

From a Hayekian perspective, two sets of cointegrating vectors are of interest:

1. Five cointegrating vectors normalized with respect to the five interest rates. These five equations indicate how employment in each of the nine industrial sectors responds to changes in each interest rate.
2. Nine cointegrating vectors normalized with respect to sectoral employment. These nine equations would indicate how changes in the different interest rates affect employment in each sector.

A positive relationship between an employment rate and an interest rate indicates a late stage of production, while a negative relationship indicates an early stage. Table 3 indicates that mining (Min.), transportation and utilities (Trns.), retail trade industry (Ret.), and wholesale trade industry (Whl.) belong to the late stages of production because of their positive signs. The negative coefficients for manufacturing (Mfg.), construction (Con.), finance, insurance, and real estate (Fin.), government (Gov.), and services (Svs.), suggest they are early stages. The obvious incoherence is that mining is, here, a late stage of production. Mulligan says that if the data is less aggregated, it would be certainly possible to better disentangle the problem. For example, some mining activities like petroleum production and field services are better conceptualized as late stages. The same production can be either early or late stage depending on whether the output is sold to a final consumer or another firm. Another possibility is that they will find easier to increase both output and labor employment, rather than to expand the infrastructure, and they will act more like late-stage producers, even while producing primarily in early stages. Mulligan further states that their behavior would be attributable to the circumstance that the primary way to expand early-stage mining operations is to construct new mines, not expand existing ones.

For manufacturing, none of the negative coefficients are significant at 5% levels. This may perhaps suggest that Mfg tends toward early stage unless it is better characterized as intermediate stage. For services, the negative coefficients are significant only for long term maturities (e.g., 3- and 5-years). Mulligan affirms it may indicate that employment in services responds only to long-term interest rates, which is not surprising for early-stage production because it is considered more roundabout. For finance, the fact that the negative signs are significant only for 3- and 6-months maturities suggest this sector responds only to short-term maturities. The contrast between services and finance, Mulligan believes, may be due to the lack of capital intensity in the financial sector. Service-sector producers maintain large and expensive capital stocks and are less free to adjust the size of their workforces. Financial, insurance, and real estate employers are less dependent on capital equipment and do adjust workforce size very quickly to accommodate changing business climates. In addition, finance employers face demand that responds more directly to short-term interest rates than demand faced by service employers.

The fact that the standard errors for wholesale are (three times) greater than for retailing may suggest that retail is better characterized as late stage than wholesale.

Mulligan summarizes his findings by stating that, although early (late) stages have seen their employment levels decrease (increase) when the interest rates go up, in times of recession the employment at every stages diminishes but it occurs faster among the early stages of production.

9. Is Austrian Business Cycle Theory Still Relevant?
(Carilli, & Dempster, 2008)

Carilli & Dempster (2008) also carried out a Granger causality test on U.S. data (1959Q1-2007Q2). These authors reviewed briefly some previous studies. They noticed some shortcomings; none of them has ever tested the ABCT prediction against other theories’ prediction. That is, these previous studies confirmed the data is consistent with ABCT but not that the ABCT provides the best explanation, against the other theories. Thus, Carilli & Dempster note with regard to Keeler 2001 study :

Monetarist theory provides an alternative hypothesis in that central bank authorities react to steeper yield curves (which signal increased inflationary expectations) by dampening money supply growth, reversing the aforementioned liquidity effects of an earlier monetary expansion and bringing a liquidity-induced boom to an end.

Unlike Keeler, Carilli & Dempster don’t use the term spread (long-term minus short-term interest rates) on the grounds that it assumes that the long-short term spread is driven solely by expectation of future short-term rates. They point out that :

… other factors, such as liquidity preference and price risk (which are themselves sensitive to monetary policy changes), seem to play an important role in the structure of interest rates. These factors, along with central bank management of the long–short spread, distort the influence of personal time preferences on the term structure. Therefore, any result based on the assumptions of a purely expectations-driven term structure and money-neutral long-term rates is suspect. [6] The important point here is that the measure of the natural interest rate should be materially independent of monetary policy actions.

They thus use two other indices of the natural interest rate(s) : 1) the real growth rate in GDP 2) personal savings-consumption ratio. The first is the mainstream specification. The second has been defended by Rothbard in “Man, Economy and State” (1962 [2009]). They described the strength of this proxy as follows :

To illustrate, consider a monetary inflation (reserve increase) that not only lowers effective market rates, but also decreases the natural rate proxy. If the latter effect is transitory, which is the case according to ABCT because the natural rate is a pure reflection of time preference, the endogenously determined lag structure will eliminate the distorting influence of these monetary effects on our proxy, so that the estimated relationships reflect the direct influences of reserves on the underlying (and unobservable) interest rate gap.

The ADF tests reveal that total reserves (U.S. (FFR) federal funds rate) and real GDP are not stationary, and are I(1). This problem is corrected through first-differencing. They then make use of the vector autoregression (VAR) and Granger causality tests to evaluate whether central bank policy (with a variable they call “reserves” through the use of the proxy FFR) can cause interest rate gap (divergence between market and natural rates) which in turn cause GDP. This causal chain is sine qua non for the tenability of ABCT. Concretely, the test is splitted in two phases :

Δ Reserves → Δ Interest Rate gap
Δ Interest Rate gap → Δ Output

Both linkages must hold for the causality to be tenable. Using the first approximation of the natural interest rate(s) they failed to reject the null hypothesis of no Granger causality with regard to Δ Reserves → Δ Interest Rate gap. However, using Rothbard’s recommendation of the natural interest rate(s) they succeeded in rejecting the null hypothesis for both linkages.

And they perform another, more thorough test of the ABCT, by answering whether or not the interest rate gap leads to an artificial expansion followed by a contraction. This is accomplished by a polynomial distributed lag (PDL) function. Concretely, in regression modeling we expect to see β0, …, βp > 0 and βp+1, …, βk < 0, i.e., the first set of intermediate multiplier must be positive and the second set to be negative. As the lag of interest rate gap increases, the (positive) coefficient must become negative, having a quadratic curve.

For each natural rate proxy, the effects of an interest rate innovation (through central bank manipulation) are at first positive and then negative on GDP. In the case of the mainstream natural rate proxy (Table 3), the negative effects are short-lived, occurring for about three quarters about 2 years after the interest rate gap appears. In quarters t-11 and t-12, the effect becomes positive again. For the Rothbardian natural rate proxy (Table 4), however, the effects become negative about ten quarters after the gap appears and remain so afterward.

By providing evidence of an endogenous turning point in the relationship between interest rates gap and output, they strengthened the relevance of the austrian theory of the business cycle versus the alternative theories.

10. Is Austrian Business Cycle Theory Still Relevant?
(Carilli, Dempster & Rasmussen, unpublished)

Carilli, Dempster & Rasmussen previously examined the case of the Japan’s lost decade (1981-1996) by way of Granger causality tests. They tested the null hypothesis of whether the japanese discount rate (money supply, M2) does not Granger cause interest rate gap. The consumption-savings ratio is chosen for proxying the gap between the market and natural interest rates (but not the mainstream specification of the natural interest rates) :

To calculate our proxy, we followed two steps. First, after determining that consumption-savings ratios were trend stationary, we detrended the ratio accordingly to obtain the long-term average (predicted) values. Second, the residuals from this detrended series were indexed to the appropriate money supply series to remove the effects of money supply variation on the ratio. The indexed series represents the natural rate (which is unaffected by variations in money supply). The interest rate gap is defined as the difference between this ratio and the actual (non-indexed) “market” ratio, which is caused to fluctuate by means of central bank-directed monetary policy.

The results were again compelling for the austrian theory. The p-value of 0.007 implies rejection of the null hypothesis. Their second test is whether or not the interest rate gap does not Granger cause nominal GDP. The p-value was 0.013. Here again, the hypothesis of no causality must be rejected.

These authors also investigated the U.S. business cycles during the 1980-1999 period. Here again, they discover that the US federal funds rate (money supply, MZM) Granger caused the interest rate gap, with p-value of 0.041. The interest rate gap also Granger caused nominal GDP, with p-value of 0.003.

A further test, again, involved the ABCT hypothesis that the movements in GDP will be upward at first only and then downward as malinvestment makes itself apparent. They seek to determine whether GDP growth has a turning point endogenous to the interest rate innovation. The interest gap is expected to have a positive impact on GDP in earlier periods and a negative impact in later periods. For both the Japan and the U.S., the expected pattern was apparent.

11. Empirical Evidence for Hayek’s Theory of Economic Fluctuations
(Charles E. Wainhouse, 1984)

Wainhouse (1984) was probably the first economist to have applied econometric tests on the ABCT. The data (all seasonally adjusted except for interest rates) used are monthly series for the period January 1959 through June 1981. He uses Granger causality to test several assumptions. The first is that :

1. Changes in the supply of savings are independent of changes in the supply of bank credit.

Concretely, Wainhouse tested the null hypotheses of 1°) Δsavings do not cause Δcredit and 2°) Δcredit do not cause Δsavings. F-tests of significance reveal that the null hypothesis of both assumptions is accepted. None of these variables Granger cause the other.

The second tested assumption was that :

2. Changes in the supply of credit lead changes in rates of interest. Furthermore, changes in credit and interest rates are inversely related.

Concretely, Wainhouse tested the null hypothesis of Δcredit does not cause Δrate of interest. The null hypothesis is rejected at both the 1% and 5% levels for fifty-five of the fifty-seven cases, (three measures of credit, nineteen interest rates) and at the 5 percent level for the remaining cases. That the causality is unidirectional has been provided by a further Granger test of the null hypothesis that the changes in interest rates do not cause the changes in credit. The null hypothesis is accepted in all of the 57 cases.

The third tested assumption was that :

3. Changes in the rate of change of credit lead changes in the output of producer goods.

Concretely, Wainhouse tested the null hypothesis of Δ% Δcredit does not cause Δoutput of producer goods. For 102 of the 120 cases analyzed, the null hypothesis is rejected. However, he noted that among 19 cases out of the 120 cases, there had been evidence of bidirectional causality. But overall, the pattern is more consistent with ABCT prediction than the reverse. The acceptation of these 3 propositions implies that credit changes causes rate interest changes which in turn causes changes in the output of producer goods, while savings do not enter into this equation. Interestingly, the author cited Hayek’s Profits, Interest and Investment (1939, p. 64) in saying that a business cycle can be triggered without any change in the interest rates because the changes in the rate of profit (due to artificial credits) can fulfill the same role.

Now, the author begins to examine the expected movements in relative prices. Six episodes of credit expansion are considered : “I” for October 1964, “II” for June 1968, “III” for September 1972, “IV” for July 1977, “A” for June 1970, “B” for September, 1976.

The fourth tested assumption was that :

4. The ratio of producer goods prices to consumer goods prices tends to rise after the initiation of a credit expansion.

The author uses BLS data on consumer price index (CPI) for all items for urban consumers, seasonally adjusted (1967=100) and the BLS data on producer price index (PPI) for a wide range of stages of processing and commodity groupings taken as measures of producer goods. Of the 162 cases examined (6 credit expansion episodes, 27 relative prices) the expected pattern is coincident with, or shows a lag with respect to, the onset of a credit expansion in 110 instances. Of the 52 cases in which the relative price displays no evidence of rising in the neighborhood (up to twenty-four months after the onset) of a credit expansion, 26 occur in cases A and B, with 19 in B. In episode B, only 8 of 27 relative prices (5 with a lag) show the expected pattern, suggesting that this episode may be misidentified as a point of credit expansion.

The fifth tested assumption was that :

5. The prices of producer goods closest to final consumption tend to decline relative to the prices of producer goods further away from the consumer good in the production scheme.

The author examined the ratio of producer goods prices located close to (PPIN) and far from (PPIF) final consumption, or ratio PPIN/PPIF. Of the 300 cases analyzed, 213 confirms the expected prediction, either coinciding with (145) or lagging behind (68) the onset of the credit expansion. Of the 87 cases in which PPIN/PPIF fails to move as expected, 59 occur in the expansion episodes A and B.

While during the early phase of the boom, the ratio CPI/PPI is expected to decline, this ratio is expected to increase during the late phase of the boom. Therefore, the sixth tested assumption was that :

6. The prices of consumer goods rise relative to the prices of producers’ goods, reversing the initial shift in relative prices.

The examination of the same set of relative prices used in the 4th assumption reveals that an expected quadratic curve is evident in 89 of the 162 cases, and of these 89, 14 show a lag with respect to the onset of the credit expansion, whereas 75 seem to be coincident. Here again, it seems, as was the case with assumptions 4° and 5°, that ABCT’s failed predictions tend to be concentrated on expansions A and B. Overall, proposition 6 appears tenable. But it must be kept in mind that the role of relative prices in consumer/producer goods has not been confirmed very well in subsequent research. Finally, it seems Wainhouse has not conducted a test for the presence or absence of unit roots, a basic requirement for time series regressions. However, the series were prefiltered for stationarity by taking first differences.

12. The Capital Structure and the Business Cycle: Some Tests of the Validity of the Austrian Business Cycle in South Africa
(Le Roux & Levin, 1998)

Le Roux (1999) has tested the ABCT in a manner very similar to Wainhouse (1984). The study focuses on South Africa’s cyclical experience from 1980 to 1996. The data consist of various credit, savings, price and investment series from the Reserve Bank’s databank. Granger causality is used. The number of observations is generally lower than what other studies have for this kind of test.

Granger-causality test reveals that 1) changes in the supply of savings are independent of changes in the supply of bank credit with no causality in either direction, 2) changes in the supply of credit lead to changes in rates of interest but not the reverse, 3) changes in the rate of credit growth lead to changes in the output of producer goods (manufacturing and secondary industry) but not the reverse.

The author then looks at the trend in the ratios of producer/consumer prices following the credit expansion. Figure 4 shows that the ratio producer/consumer goods prices increases after the initiation of credit expansion. Increases (decreases) in the rate of growth of credit correspond with the upswings (downswings) in the business cycle. Figures 5-8 show that, after credit expansion, the prices of goods closer to final consumption tend to decline relative to the prices of producer goods further away from consumer goods. Figures 5-8 also shows that the prices of these consumer goods rise relative to the prices of these producer goods towards the end of the credit expansion.

13. Evidence Regarding the Structure of Production
(Larry J. Sechrest, 2004)

Sechrest (2004) provides some additional but weak support for the ABCT. The weakness of the analysis is to be merely correlational, not causational, although the pattern of correlation is coherent with predictions endorsed by ABCT. The data in use covers US period of January 1959 (the base period=100) to December 2002 (all the data is monthly and not seasonally adjusted, for a total of 528 observations). He compares the regressions of 3 monetary bases, M1, M2 and the austrian measures of money supply, which is defined as follows :

As suggested by Salerno (1987: 1-6; 2004) and Rothbard (1978: 143-56; 1983: 254- 62), the Austrian conception of money is here taken to include currency held by the public, demand deposits, other checkable deposits, U.S. government deposits, and deposits due to foreign banks and foreign official institutions. Ideally, overnight repurchase agreements, overnight Eurodollar accounts, and U.S. savings bonds should also be included but are excluded due to the fact that government data 1) combine term repurchase agreements and term Eurodollars with the overnight categories and 2) no longer include numbers for the stock of savings bonds.

Despite the inherent problem of R² (adjusted or not for the number of independent variables) in that it systematically reduces the real effect size, the R² in his series of regression were always very high. But the more relevant piece of information is the relative strength of these correlations (preferably, in terms of r but not r²). For example, in the three following dependent variables, 1) the total industrial production, 2) business equipment production, 3) commercial and industrial loans, the correlation of savings rate (around r=0.70) with either of these variables is less than money supply growth (around r=0.95). Among the 3 considered money supply measures, M2 has the highest correlation, followed by austrian measures and M1, but the difference is just meaningless in terms of r. Anyway, all this suggests that savings rate has less explanatory power than money supply in the prediction of production and commercial loans.

Next, Sechrest performs regressions with the following 3 independent variables : the ratio CPI/PPI (i.e., ratio of consumer goods to producer goods prices), the difference between Federal Funds rate and a proxy for natural interest rates (approximated by the 3-month lagged difference between consumer and producer prices), and finally, money supply (either M1, M2, or austrian measures). The dependent variable is, firstly, total industrial production, and secondly, production of business equipment; in both cases, the model R² are similar and larger than 0.90. Although the R² for these 3-variable models is higher for M2 and austrian measures, the difference is meaningless in terms of r. Using the same dependent var., but replacing money supply by commercial and industrial loans leads to a modestly higher R² but a meaningless improvement in r.

Finally, Sechrest computes three averages of prices : 1) COMPIND1, the equally weighted average of CPI and PPI, 2) COMPIND2, the equally weighted average of CPI, PPI, the Dow Jones Industrial Average and urban real estate rental prices, 3) COMPIND3, the equally weighted average of CPI, PPI, Dow Jones Industrial Average, urban real estate rental prices and wage rates paid to private-sector, goods-producing workers. The R² of M1 were (respectively) 0.946, 0.892, 0.899. The R² of M2 were (respectively) 0.952, 0.946, 0.954. The R² of austrian measures were (respectively) 0.893, 0.947, 0.954. In other words, by adding more categories of prices, the correlation declines for M1, remains constant for M2, and increases for austrian measures. Again, in terms of r, either the difference or increase is meaningless.

14. Austrian Business Cycle Theory and Global Financial Crisis: Some Lessons for Macroeconomic Risk and Financial Stability
(Ersan Bocutoğlu & Aykut Ekinci, 2010)

Table 1 shows the relationship between Industrial Production Index (IPI) growth and lagged (lag) and future (lead) term spread (the difference between long and short rates). The correlations indicate that, in general, the higher the spread is (that is, the more steeply sloped the yield curve is) and the higher the rate of future IPI growth is.

Figure 4 shows CE/IE ratio and CM/IM ratio, the Crude energy materials divided by Intermediate energy goods and Crude nonfood materials less energy divided by Intermediate materials less food and energy. The logic is that the ABCT expects relatively higher price fluctuations in the first stages of production compared to later stages, and the picture shows indeed greater ratio in the periods before shaded areas (recessions).

Figure 5 displays the COMPOSITE Index obtained from the equal combining of the two index series mentioned above. The index shows large increases/declines in the period of booms/recessions.

From 2001 to 2007, the CPI, and industrial-inputs, energy, and metal prices increased by about 19%, 113%, 174% and 226% respectively. In the same period, an increase of more than 250% occurred in the Global Dow Jones Index.

As we see, the CPI shows no anomalous trend (ups and downs) during the whole period. CPI inflation was 3.5% during 1981-2001 but 2.9% during 2002-2007. Selgin (1997, pp. 56-57) has already noted that inflation has nothing to do with the business cycle, as the prices did not even rise in the 1920s boom.

VIX Index is the Chicago Board Options Exchange (CBOE) Volatility Index, which shows the market’s expectation of 30-day volatility. This volatility is meant to be forward-looking and is calculated from both calls and puts. The VIX index is a widely used measure of market risk and is often referred to as the “investor fear gauge.”

One variable rises in value at the same time the other variable declines in value. The negative correlation between SPREAD and risk perception is illustrative of the proposition that an artificially reduced market interest rate (due to credit expansion) lead to underestimation of risk perception in financial markets.

The volatility of CPI and IPI were measured as the percentage deviation of index from the Hodrick-Prescott trend. We computed this standard deviation by using a rolling window of twelve months. Volatility of output and inflation lagged behind spread.

15. Financial Market Shocks during the Great Depression
(Alycia Chin & Missaka Warusawitharana, 2010)

16. The Recession of 1990: An Austrian Explanation
(Arthur Middleton Hughes, 1997)

Hughes (1997) describes briefly the leading consequence of money over-expansion. At some point, interest rates must go up, tending to their “normal” levels :

As interest rates go up, new projects not yet started will be canceled. But many of those which are only half finished will also have to be abandoned. One reason is that capital financing is often obtained on a pay-as-you-go basis. As industries compute the payoff for a project started when interest rates were 8 percent, which now must compete for funds at 12 percent, they realize that the project is a loser. They cut their losses, and abandon the enterprise. The workers are laid off, and often, much of the project is a total loss. The reason? Because most capital goods (semifinished goods and facilities) are specific to an industry and have little general usefulness. … As workers are laid off in higher-stage industries, they reduce their spending for consumer goods. The recession spreads.

Of couse, the workers, unlike semi-finished goods, can be specialized but are not themselves specific to any industry. Unemployment problems can still be corrected to some extent. This aside, the money supply growth averaged 9.6% per year from 1981 to 1986 but only 4.1% from 1987 to 1991. The figures for GNP were respectively 2.6% and 2%. Hughes expects the following : since the more capitalized industries expanded the most at earlier years during the boom, more workers are demanded and hired by them, which resulted in pay increases. These workers would spend more money on consumer goods (to the extent that their time-preference did not changed) the lower-stage industries react by increasing bank credits. This, obviously, occurs after the previous increases in bank credits by higher-stage industries. Given this, the lower-stage industries expansion is based more on anticipated consumer demand than upon the availability of capital, as was the case for higher-stage industries.

That the demand for bank credit came earlier in the more capital intensive industries is represented by iron and steel, primary metals, and machinery industries in Figure 3 which depicts a relatively large increase in long-term bank loans for higher-stage industries compared to “all manufacting bank loans” at index 100. Hughes explains why the expansion wasn’t due to increases in private demand :

The fact that the increases in higher stage borrowing from 1981 to 1985 were based on the increased availability of capital funding rather than directly anticipated demand for increased output is shown by the statistics on industrial production during this period. The level of production in the iron and steel industry at the time was far below capacity (estimated at about 63.5 percent from 1981 to 1985), due to the low level of orders. Lower-stage industries, closer to the consumer, showed no such increased borrowing levels during this period of massive money supply increases.

With the benefit of hindsight, we know today that net shipments of steel mill products would never regain their 1981 levels during the following decade. Why did they expand their production? All that industry participants knew at the time was that funding for expansion was available. Iron-ore production, much closer to its direct customers than the iron and steel industry, did not invest in expansion during the period, and, in fact, closed down several of its operating mines.

When the massive money-supply increases came to an end in 1986, the iron and steel industry collapsed. Prices of their product dropped every year as the higher-capacity and more-efficient new facilities competed with the older plants for what was essentially a disappointing demand.

Figure 4 depicts the long-term bank borrowing of three lower-stage industries during 1981-1991 compared to manufacturing industries during credit expansion from 1981 to 1984. Rubber, Food, and Textiles industries peaked at 1986 and 1987 but even after the drastic reduction in money supply growth starting from 1987, their long-term bank loans had not decreased in 1991. According to Hughes :

The previous expansion in higher-stage industries produced an increase in consumer spending which pushed lower-stage manufacturers to borrow to expand their facilities to meet it. [16] … From 1981 to 1986, real consumption spending in 1987 dollars increased by 19.7 percent, whereas real GNP increased only 15.7 percent in the same period. Consumers were trying to restore their spending patterns after the recession of 1982.

In Figure 5, we see Iron and Steel expanding the most during the period of massive money growth whereas the Food industry starts its large expansion only in 1987. More specifically :

Figure 5 puts the entire period into perspective. It shows total dollar borrowing by all manufacturing industries adjusted for inflation – the producer’s price index. … The dramatic reduction in money supply increases in 1987 did not affect the total borrowing level at all – in fact the average annual increase in borrowing after 1986 was higher than before 1986. What did change was the distribution of that borrowing. …

Most traditional economists looked on the years from 1986 through 1988 as being boom years. They overlooked what was happening in higher-stage industries. What was going on in iron and steel, for example was this: LTV Corp., Wheeling-Pittsburgh Steel Corp. and Sharon Steel Corp., were all forced to file for bankruptcy following the collapse of steel demand in 1985 and 1986. This collapse accelerated the reduction of U.S. steel-making capacity, and triggered a major restructuring of the iron ore and steel industries on both sides of the US-Canadian border.

The steel (and copper) industry expanded so that their inventories increased drastically by 275% of 1979 levels in 1983. Since 1984, and despite this build-up, inventories fall dangerously every year until 1988, and reach 14% of 1981 levels by 1989. The large fall in prices for Iron & Steel starting from 1986 (Figure 6) coincided with the Fed having stopped inflating money supply. Data shows that world consumption of copper was flat between 1980 and 1989. So its expansion in early 1980s can’t be ascribed to consumption demand.

That higher-stage (as opposed to lower-stage) industries expansion coincided much more with credit expansion is best illustration by the following :

In figure 7, we contrast long-term borrowing of all manufacturing, retailing, and wholesale firms. Retail borrowing took a nosedive from 1981 to 1982 – recession years – and stayed down all during the period when the Federal Reserve was inflating the money supply. Retail borrowing only accelerated in 1987 – after the inflation of the money supply was over! Why? Because retail firms borrow to meet immediate customer demand. Higher-order firms borrow when financing is available on attractive terms. … During all of these years, prices were flat or falling. Then, in 1986, prices began to climb.

In the later stages of the boom, consumer spending competes with and overtakes all other types of activity. It is at this point that unused higher-stage capacity materializes because, as Hayek says, “We are unable to use the fixed plant to the full extent because the current demand for consumer’s goods is too urgent to permit us to invest in current productive services in the long processes for which (in consequence of ‘misdirections of capital’) the necessary durable equipment is available.”

Cwik (1998) commented on Hughes (1997) study and he observed an obvious mistake in Hughes presentation of the ABCT fundamentals because the theory does not claim that highly intensive firms will make investment of longer duration (i.e., that they need more time to build their products) even though Cwik agrees that higher-stage firms are more sensitive to interest rates and consequently have wider swings in a cycle, as was also noted by Skousen, cite by Huerta de Soto (1998, [2009], pp. 379, 420-421, 502). Cwik also points out that monetary policies may not act to end the boom, and yet it will necessarily come to an end simply because of shortages of real resources, which cannot satisfy increases in both investment goods and consumer goods.

17. The recession and Austrian business cycle theory: An empirical perspective
(William N. Butos, 1993)

Butos (1993) has previously studied the 1980s recessions. But more generally, his data covers the periods 1970 to 1992. All data was seasonally adjusted. He begins to relate the trends in bank reserves/credit from 1973 to 1993.

As Figure 2 shows, commercial bank’s industrial and commercial loans increased steadily until the end of 1989. Although these loans fall off rather sharply beginning in 1990, which is consistent with the ATBC, until that time they failed to expand as vigorously as might be expected given the increase in bank reserves. Several factors may explain this result.

First, the rapid increase in bank reserves relative to bank credit reflects the Fed’s apparent desire to offset the strong decline in the income velocity of money, defined as the ratio of nominal GDP to the stock of money, which occurred between the end of 1983 and the end of 1986. Income velocity will decline if individuals increase their holdings of money relative to their incomes. This presumably is what happened as falling interest and inflation rates reduced the opportunity cost of holding money, and as an expanding array of interest-earning money assets became more widely available following the 1980 Depository Institutions Deregulation Monetary Control Act (DIDMCA) and the 1982 the Garn-St Germain Depository Institutions Act.

Second, the third-world debt crisis induced banks with third-world loans to substantially increase their loan loss reserves – reserves set aside (not loaned out) to cover nonperforming loans. …

A third factor concerns the dramatic developments in the real-estate market during the 1980s. The data suggest that a significant element of the 1982-1990 expansion was real estate related. As Table 1 shows, in the 1980s commercial banks increased their real-estate loans at more than double the average rate of the over cyclical upswings. An important factor in accounting for the growth in real-estate loans was the 1981 Economic Recovery Tax Act (ERTA). ERTA provided substantially more favorable tax treatment of investment by accelerating depreciation schedules (15 years under ERTA compared to 40 years under pre-ERTA); reducing the asset lives of plant, equipment, commercial buildings, and rental housing; and providing a more generous investment tax credit on equipment (but not structures).

That the economic boom is followed by a decline in real (i.e., inflation-adjusted) interest rates and the bust being associated with increased real interest rates is consistent with ABCT, although the impulse originates from monetary expansion, not necessarily interest rates (see here). In any case, Figure 3 shows the trends in nominal interest rates for short- and long-term rates (respectively, 3-Month Treasury Bill Rate and AAA Bond Yield). Short-term rates fluctuate more, increasing dangerously during the recessions, i.e., 1973-1975, 1980-1982, and 1989-1990. Figure 4 shows that the real long-term rates follow a path not entirely consistent with ABCT; its sharp increase from 1980 to 1982 cohere with ABCT but not its decline during the 1973-1975 recession. As for both of these paths, Butos speculates that the inflation uncertainties from early 1970s to early 1980s could be the explanation. Its because the real interest rate(s) is defined as nominal (or market) rate minus expected rate of inflation. The real rate is not an observable variable unlike the market rate and it thus requires to make claims about investors’ expectations of inflation. The 1970s period is indeed very particular due to its associated stagflation and the existence of negative real interest rates in the mid-1970s. Butos concluded on this :

In any case, there are clear pitfalls in making strong inferences based on the movement of calculated real interest rates. This is especially the case when the purchasing power of money is unstable, as has been the case during the past two decades. Inflation, because it creates unsystematic and unpredictable changes in relative prices, interferes with their informational role, thereby making economic calculation difficult. This applies with special force to “time markets,” in which the efficacy of current decisions is decisively linked to a future made even more uncertain by inflation.

Information about the employment structure is provided. Given Figure 5, it can be seen that the index of durable goods production shows more fluctuations during the periods of recession (i.e., 1973-1975, 1981-1982, 1990-1991). And Figure 6 gives another strong support for the ABCT story because the index of capacity utilization in 1982-1983 and 1991 experienced a much greater decline among the durable manufacturing.

Figure 7 shows there is no large fluctuation in the ratio of total private nonagricultural employment to employment in durable goods-producing industries during all contractions since 1973. In comparison, the ratio nonagricultural/nondurable goods industries employment increased during the recessions around 1973-1975, 1981-1983, but not in the 1990-1991 recession. And yet, according to Figure 8, there had been a large decline in employment among the manufacturing industries during that period, as depicted below :

In general, durable goods industries are more sensitive to the business cycle than are nondurable goods. With regard to Figure 8, Butos noted :

… between 1989 and 1991 employment in durable goods industries fell by 10.4 percent, while only by 2.2 percent in nondurable goods industries.

According to the ABCT, the new orders and shipments of capital goods should decline during recessions due to reallocation of resources back toward less capital-intensive production methods. This is illustrated in Figures 9 & 10, where new orders and shipments tracks overall economic activity. New orders have declined during the recession periods, 1973-1975, and 1981-1982, and finally 1990-1991 whereas the evidence is mixed for new shipments. Finally, Figure 11 shows that the ratio PPI/CPI, or producer price index to consumer price index increases at the onset of the recessions, following paths not necessarily consistent, or just weakly so, with ABCT. On the contrary, we expect the ratio to diminish during the recession, as to reflect the adjustment process of resource reallocation back toward less roundabout methods of production due to higher rates of profit in consumer goods industries relative to capital goods industries. As noted in the above econometric analyses, the ratio PPI/CPI does not have any strong explanatory power with regard to the business cycle anyway.

18. Does Austrian Business Cycle Theory Help Explain the Dot-Com Boom and Bust?
(Gene Callahan & Roger W. Garrison, 2003)

Callahan & Garrison (2003) relate the story of the dot-com bubble (1995-2002). They indicate that between 1992 and 2000, the personal savings went down from 8.7% to -0.12% whereas it fluctuated always between 7.5% and 10.9% during the period of 1950-1992. The 1992-2000 period of lower saving rates coincided with an increase in households’ debt (in proportion of personal disposable income), with 80% in the second half of the 1980s and 97% in 2000. Businesses, at the same time, and probably in reaction, increased their capital spending : “For instance, a paper by several Northern Californian organizations reports that venture capital firms invested \$50.7 billion in the Bay Area during 1999–2000, which was \$35 billion, or 233 percent, over the trend-line increase (Bay Area Council et al. 2002, p. 12).” (p. 85). The decline in savings has been accompanied by an increase in both consumption and investment. Without savings to support them, a subsequent bust is expected.

During what we will roughly designate as “the boom,” from June 1995 to March 2000, MZM (money zero maturity, which represents all money in M2 less the time deposits, plus all money market funds) grew 52 percent, well ahead of real GDP growth of 22 percent (Rogers 2002) for the same period. The interest rate on 10-year Treasuries declined from 6.91 percent to 4.53 percent in October 1998, before beginning to rise again. Rates peaked in early 2000, roughly corresponding to the end of the boom. Corporate Aaa bond yields declined from 8.46 percent at the beginning of 1995 to 6.22 percent in at the end of 1998. …

By late 1999, production of business equipment was up 74 percent and construction up 35 percent over 1992, while production of consumption goods had risen only 18 percent. Among manufacturing goods, durable good production had risen 76 percent while nondurable good production had risen just 13 percent (Federal Reserve 2000). “Annual borrowing by nonfinancial corporations as a percentage of nonfinancial corporate GDP darted from 3.4 per cent in 1994 … to a previously unparalleled 9.9 per cent in the first half of 2000. … As a result, by the first half of 2000, nonfinancial corporate borrowing on an annual basis had more than quadrupled with respect to 1994 and nonfinancial corporate debt as a proportion of nonfinancial corporate GDP had reached 85 per cent, the highest level ever” (Brenner 2002, p. 192). …

Even as low interest rates spurred investment in certain capital goods, they led to a collapse in savings. The personal savings rate declined from an already low 2.1 percent … in 1997 to -1.5 percent by 1999 (Bureau of Economic Analysis 1999). Consumers were increasingly leveraged, especially on their homes. “In 1989, about 7 percent of new mortgages had less than a 10 percent down payment, according to Graham Fisher & Co., an investment research firm. By 1999, that was more than 50 percent” (Priest 2001). …

Of particular interest is the increase in salaries in dot-com related jobs, and with it, some accounts that people deliberately changed their jobs in order to take advantage of the highly paid wages in dot-com jobs. The ABCT predicts relatively larger wage rate increase among jobs in areas where the credits had been concentrated the most. Also, there had been an increase in apartment rents, e.g., in San Francisco, from \$920 per month in the fall of 1995 to \$2,080 in the spring of 2000, presumably due to the increase in demand for housing, stemming from the influx of wealthy dot-commers.

The people needed to staff dot-com companies were also rapidly becoming more expensive. As the boom peaked, Audi (2000) reported: “So many San Francisco lawyers were leaving good jobs in big firms to work for start-up Web companies that to compete, some firms doubled the starting pay to \$150,000.”

Kuo (2001, p. 46) tells of how, within weeks of being hired in the summer of 1998, Value America executive Glenda Dorchak demanded, and received, “a lot more stock … a significant raise in pay,” and a promotion.

Covin (2002) noted: “In the late 90s, there was a sudden increase in programmer salaries as a result of the dot-com boom. Programmers who were earning \$45,000 in 1995 were making well over \$100,000 by the year 2000.”

Even Microsoft had to increase salaries. Companies that rely on credit enhance competition with other competitors for access to factors of production, such as workers.

19. Can Austrian Theory Explain Construction Employment?
(Robert P. Murphy, 2011)

As the housing boom intensified, sucking more and more workers into construction, the national unemployment rate steadily fell. Then, as the housing boom tapered off, extra workers stopped getting siphoned into the housing sector, and the national unemployment rate bottomed out. Finally, as construction employment began falling, the national unemployment rate began rising.

Now DeLong or Sumner might object that I’m mixing up causation with correlation. For whatever reason, consumers got yellow in late 2007 and did the unthinkable — they started saving some of their incomes. Thus, everything started dropping at that point, including home prices and construction employment. But there’s nothing “real” in this story corresponding to the Austrians’ worries about “too much housing,” Sumner and DeLong might argue.

20. Putting Austrian Business-Cycle Theory to the Test
(Robert P. Murphy, 2010)

In the Austrian framework, construction would typically be the “highest order” of these, because things like office buildings and houses are very capital-intensive and provide a flow of services for decades. Next in line would be durable-goods manufacturing, while nondurable-goods manufacturing would be the “lowest order” of these three categories.

21. Money, Bank Credit, and Economic Cycles
(Huerta de Soto, 1998 [2009])

Huerta de Soto (1998, pp. 487-490) relies on Rothbard’s account of the roaring twenties (1920s). The money supply in the United States grew from \$37 billion in 1921 to over \$55 billion in January 1929. Robert P. Murphy (2011) says that the 1920s have experienced slight declines in prices but that such event did not cause recession and did not prevent economic growth. Selgin (1997, pp. 56-57) agrees that price stability has masked growing profits and market instability. Selgin has noted that wages (and costs) could not catch up with firms’ revenues, an illustration of sluggish adjustment of wage-rates. The ABCT predicts this will happen because the rise in profits is always ahead of the rise in costs, to the extent the money supply keeps increasing. Huerta de Soto said that the price of securities increased four-fold in the stock market. The production of goods for current consumption grew by 60% throughout the period while the production of durable consumer goods, iron, steel, and other fixed capital goods increased by 160%. More importantly, during the 1920s, the wages rose mainly in the capital goods industries by 12% compared to the 5% in consumer goods industries.

Huerta de Soto (1998, p. 502) cited the works of Skousen (unpublished) and Ramey (1989) regarding the fluctuation of prices during the cycles. From Skousen, we know that during the period 1976-1992, the prices of products from the stages furthest from consumption varied from +30% to -10%, the prices of intermediate goods only oscillated between +14% and -1%, and the prices of final consumer goods varied from +10 to -2%. The fact that the more capital-intensive industries have expanded and contracted the most is well within the prediction of the austrian theory. From Ramey (1989), we discover that the inventories closest to the final consumption stage show most stability. Ramey (1989) examined the seasonally adjusted quarterly data from 1960 to 1984 regarding several durable-goods manufacturing. Table 7 shows the own-price elasticities for materials “m”, goods-in-process “g”, and finished goods “f” as well as labor “w” in four industries, i.e., primary metals, fabricated metals, nonelectrical machinery, electrical machinery, estimated for 1984. In each industries, the own-elasticities in the diagonals shows that materials have less elasticities, goods-in-process have more elasticities, and finished goods the most. Table 8 shows the output elasticities estimated for 1984. Materials constantly have the smallest elasticities in all four industries. However, it seems that goods-in-process have more elasticity than finished goods. The fact that the higher-order goods tend to be less stable over time is a sign that they are more sensitive to interest rates and monetary policies. This is also predictable from the ABCT.

3. References.

1. Anker, R. (2011). Austrian Business Cycle Theory Evidence from Scandinavia. Dissertation.
2. Bismans, F., & Mougeot, C. (2009). Austrian Business Cycle Theory: Empirical Evidence. The Review of Austrian Economics, 22(3), 241-257.
3. Bocutoğlu, E., & Ekinci, A. (2010). Austrian Business Cycle Theory and Global Financial Crisis: Some Lessons for Macroeconomic Risk and Financial Stability.
4. Butos, W. N. (1993). The Recession and Austrian Business Cycle Theory: An Empirical Perspective. Critical Review, 7(2-3), 277-306.
5. Callahan, G. & Garrison, R. W. (2003). Does Austrian Business Cycle Theory Help Explain the Dot-Com Boom and Bust?. The Quarterly Journal of Austrian Economics, 6 (2), 67-98.
6. Carilli, A. M., Dempster, G. M., & Rasmussen H. (unpublished). Is Austrian Business Cycle Theory Still Relevant?.
7. Carilli, A. M., & Dempster, G. M. (2008). Is the Austrian Business Cycle Theory Still Relevant?. The Review of Austrian Economics, 21(4), 271-281.
8. Chin, A., & Warusawitharana, M. (2010). Financial market shocks during the Great Depression. The BE Journal of Macroeconomics, 10(1).
9. Cwik, P. (1998). The Recession of 1990: A Comment. Quarterly Journal of Austrian Economics, 1(2), 85-88.
10. Hughes, A. M. (1997). The Recession of 1990: An Austrian Explanation. The Review of Austrian Economics, 10(1), 107-123.
11. Keeler, J. P. (2001). Empirical Evidence on the Austrian Business Cycle Theory. The Review of Austrian Economics, 14(4), 331-351.
12. Keeler, J. P. (2001). Relative Prices and the Business Cycle. Unpublished manuscript.
13. Le Roux, P. & Levin, M. (1998). The Capital Structure and the Business Cycle: Some Tests of the Validity of the Austrian Business Cycle in South Africa. Journal for Studies in Economics and Econometrics, 22(3), 91–109.
14. Mulligan, R.F. (2002). A Hayekian Analysis of the Term Structure of Production. Quarterly Journal of Austrian Economics, 5(2), 17-33.
15. Mulligan, R. F. (2005). The Austrian Business Cycle: A Vector Error-Correction Model with Commercial and Industrial Loans. Journal of Private Enterprise, 22(1), 51-91.
16. Mulligan, R. F. (2006). An Empirical Examination of Austrian Business Cycle Theory. The Quarterly Journal of Austrian Economics, 9(2), 69-93.
17. Murphy, R. P. (2010). Putting Austrian Business-Cycle Theory to the Test. Mises Daily.
18. Murphy, R. P. (2011). Can Austrian Theory Explain Construction Employment?. Mises Daily.
19. Powell, B. (2002). Explaining Japan’s Recession. Quarterly Journal of Austrian Economics, 5(2), 35-50.
20. Ramey, V. A. (1989). Inventories as Factors of Production and Economic Fluctuations. The American Economic Review, 338-354.
21. Sechrest, L. (2004). Evidence Regarding the Structure of Production: An Approach to Austrian Business Cycle Theory.
22. Selleby, K., & Helmersson, T. (2009). Empirical Testing of the Austrian Business Cycle Theory: Modelling of the Short-run Intertemporal Resource Allocation. Dissertation.
23. Wainhouse, C. (1984). Empirical Evidence for Hayek’s Theory of Economic Fluctuations. In: Siegel, B. (Ed.) Money in Crisis, pp. 37–71. San Francisco: Pacific Institute for Public Policy Research.
24. Weber, C. (2009). An Empirical Application of the Austrian School’s ‘Stages of Production’. Working paper, New School for Social Research.
This entry was posted in Economics and tagged . Bookmark the permalink.