Printer Friendly

The role of credit in the 2007-09 Great Recession.


A number of researchers have attempted to analyze the causes and policy lessons of the Great Recession of 2007-2009, which has been compared to the Great Depression in a number of respects. This paper contributes to that literature. In this section, we begin with a brief methodological introduction as well as a selective review of the relevant literature on the financial crisis and the accompanying recession.

The analysis of macroeconomic data has often suffered from the fact that a preconceived theory is presumed to be true, and the statistical task is merely to confirm the preconceived theory (Campos et al. 2005). This has often been the practice among the macroeconolnists who have adopted Real Business Cycle (RBC) to be the underlying reality. RBC was later christened "stochastic dynamic general equilibrium" (SDGE). The SDGE approach also dominated a number of the research divisions of the central banks of western developed countries, including the research division of the Bank of Canada. Recently it has become apparent that such representative equilibrium models have no room for central bank intervention. Therefore, the successor models to the SDGE models are called "New Consensus Macroeconomics", which appeared in the aftermath of Woodford (2003) (See Woodford 2009). In the New Consensus, some form of a Taylor Rule is incorporated as an explicit equation of the model. However, the basic New Consensus model remains an untested theory, and the econometric estimation of it is just "clothing". (in the sense used in Campos et al., op cit.) The inadequacy of empirical macro econometrics has led a number of researchers to go back to some of the guiding principles of econometric practice articulated some time ago by Haavelmo (1944). Subsequently, a rich literature now serves applied econometric practitioners. This literature includes some influential papers, such as those by Davidson et al. (1978), Hendry and Richard (1983), Hendry (1987), and others too numerous to mention (see Campos et al. 2005 for a full bibliography). The point here is not to try to retrace the history of sound econometric practice for macroeconomics, which is now well known, but to provide a brief methodological justification for the macro hypotheses tested and the approach taken in this paper, which is the cointegrated Vector Autoregression (VAR) approach. This is a purely statistical approach, designed to assess which macro theory is supported, when no a priori macro theory is imposed.

The central topic of this paper explores what key variables should be used to account for the business cycle downturn that occurred in the U.S. in 2007-9, which led to reverberations across the global economy. For such a task, no overarching economic theory is proposed in this paper and no such model is imposed. Instead, we have chosen the cointegrated VAR approach because it offers a potentially richer way of trying to capture short-run and long-run relationships, interactions, and feedbacks which we might expect and wish to test for statistically. The fecundity of the VAR as an investigative technique has been adequately demonstrated by now, but has been eloquently restated by Juselius (2006). Secondly, the VAR approach is less pretentious about a prior role of a theoretical model. It enables one to take a time series of some chosen data (after careful sitting using statistical methods) and use VAR to search for structures in the data for not one but a number of possible relevant theories. The approach could then lead to empirical support for some hypotheses, for example about the causal directions of a small number of variables, some not previously shown to be relevant in published macro literature. Thus, we hope to generate some new hypotheses about the integral macroeconomic relationships among the variables chosen here. These new hypotheses can then be tested against new data to test for either their generality or to narrow down the domain of validity of these hypotheses. For example, the relationships uncovered in this paper between disposable income, revolving credit, profits, and aggregate spending in the U.S. data should be tested on data from the UK, Canada, and other developed countries who have all been affected to a great extent by the international transmission of the financial crisis in the U.S. as shown in IMF (2009).

It is not surprising that there are already a number of research papers that have begun the task of examining the major financial crisis that started in the U.S. in August 2007, which led to severe reverberations throughout the world, second only to the impact of the Great Depression. Naturally, some researchers have attempted to explain the crisis, draw parallels with the Great Depression, or offer macroeconomic policy lessons. It might be instructive to begin with one (only one) example of the SDGE type of approach, in which an implied but untested theory provides a straitjacket which is then dressed in econometrics. This is the attempt by Barrell et al. (2008) who use a stochastic dynamic general equilibrium model of the UK National Institute of Economic and Social Research, to which a financial sector is grafted in order to elicit the impact of future financial controls on economic growth. Since an SDGE Model is a representative agent model, which is by definition a convex model, it is unclear how informational asymmetries (which played a major role in the financial crisis) or any other externalities can be incorporated in the model without violating the fundamentals of the model itself. Not surprisingly, no interesting results were obtained from the model. Their main conclusion for tighter bank regulation in their paper seems completely unconnected to the SDGE model.

However, there is a more interesting literature that does have something useful to say about the financial crisis and its consequences. Reinhart and Rogoff (2008) amass data for six centuries (spanning 66 countries) of banking crises and debt defaults and see very little that is new in the 2007-09 crisis. Michael Bordo (2008) also puts the 200709 crisis into historical perspective of earlier big international financial crises which were triggered by events in the U.S. financial system. He cites examples of the crises of 1857, 1893, 1907, and 192%33. The 2007-09 crisis has many similarities to those of the past but also some important differences. However, he does mention one important institutional feature, namely the repeal in 1998 of the Glass-Steagall act which had separated commercial from investment banking. Following the repeal of the act, competition between commercial banking and investment banking led to a race to achieve comparable profitability through comparable leveraging. Gorton (2008) and Geanakoplos (2009) get nearer to the heart of the problem as they identify informational asymmetries between buyers and sellers on the newly securitized assets. This asymmetry is of course an externality, which points to the failure of the standard Chicago School assumptions of convex environments, continuous and complete, rational, and efficient markets and with complete information. In such a convex environment, failures and crises are impossible. The identification of informational asymmetries is a clear recognition of a form or market failure, or more generally, the existence of an externality that the Chicago School fails to understand or chooses to ignore.

On the other hand, Steven Cecchetti (2008) asks some important questions on how to deal with such externalities, even though he acknowledges that we do not as yet have clear answers. His questions are pertinent to developing future macro coordination policies. He asks: What should policymakers do when prices of leveraged assets boom? How should central banks respond to declines in the price of risky assets, and the associated increase in risk premia? Should monetary policymakers react to illiquidity in the market for specific assets, and if so how? When a highly leveraged and complex financial institution experiences losses, what is the central bank's responsibility? Should a central bank take credit risk in its lending operations, or should this function belong to the U.S. Treasury? While these are good policy questions, they do not go anywhere close to a structural explanation as to what went wrong and how the financial crisis became so severe as to engulf the whole world. Finally, Edward Leamer (2007) and John Taylor (2009) see the crisis as being caused by too loose a money supply and a failure on the part of the Federal Reserve to follow the Taylor rule in setting interest rates. Taylor also produces a counterfactual argument that had the Fed followed the Taylor rule in setting interest rate policy, the housing boom and bust would not have occurred. Learner (2007) goes further in arguing that housing is the business cycle and some version of the Taylor rule would have avoided it.

While this literature throws light on important aspects of the financial crisis and business cycle downturn, a fuller account of the institutional features that accompanied the downturn and the role of leveraging (which is in fact credit plus collateral) is needed to increase our understanding of what happened. This paper is an attempt to extend that literature. We do this by selecting a few key variables and attempt to capture short-run and long-run relationships, interactions, and feedbacks which we might expect and wish to test for statistically. We carry this task out by fitting a Vector Error Correction model to U.S. data for the period 1975 to 2007. We believe that the paper generates some new inter-relationships and interesting hypotheses which should be tested on the data for other developed countries.

The Role of Credit in a Dynamic Model

The slow but gradual process of repealing legislative acts (such as the Glass Steagall act of 1933), which were designed in the first place to monitor and maintain checks and balances in the financial markets, can be seen as a factor behind the financial crisis. This process of financial de-regulation began in 1980 with the Depository Institution Deregulation and Monetary Control act, followed by the Tax Reform act of 1986, and The Gramm-Leach-Bliley act of 1999, all of which relaxed controls on the financial markets. These and other legislative acts encouraged innovation in debt instruments and new financial tools which were high risk and eventually proved costly to the U.S. economy. Real income growth over the period 2000 to 2007 was highly skewed, with the result that aggregate spending was maintained largely through cheap credit. The cheap credit was made possible by the Federal Reserve maintaining liquidity in the markets by adopting the interest rate rule in 1995 and effective demand was maintained mainly through the growth of credit. Even large reductions in corporate and capital gains taxes and income tax cuts under the Bush Administration (1) were not enough to maintain the expansionary phase once credit collapsed.

The exponential growth of credit as the main engine of prosperity was made possible by adding new forms of debt instruments in financial markets. Dore and Singh (2009) and Singh (2008) have shown that disposable income for the bottom 80 % of the U.S. population had been declining since 1984. Only the top 20 % had increasing disposable income. In fact, more than 60 % of household earned less than $50,000 in 2007 compared to less than 10 % earning over $200,000 (Dore and Singh 2009). From 1984 to 2007, only households from the top quintile had more current disposable income than their current expenditure. Consumption for the majority of the population was therefore maintained and encouraged through the availability of credit from banking and financial institutions. Dore and Singh (2009) highlighted the importance of credit in maintaining retail sales from 1992 to 2007 in a maximum likelihood regression estimation where retail sales were regressed on revolving credit, non-revolving credit, and disposable income. The authors found that revolving credit and non-revolving credit together accounted for approximately 50 % of retail sales while disposable income only influenced 2 % of sales. Below, we investigate the above result further in a dynamic framework.

To expand on the previous work, we specify a dynamic Vector Error Correction Model in which the role of credit is explicitly taken into account. Consider the following model:

[DELTA][y.sub.t] = [a.sub.0] + [p-1.summation (over i=1)] [[PSI].sub.i][increment of y.sub.t-1]+[PSI][w.sub.t]+ [[mu].sub.t] (1)

Where),t is a 4 x 1 vector of jointly determined variables which are corporate profits, revolving credit, disposable income, and retail sales (used as a proxy for aggregate spending), [DELTA] is the differencing operator, such that [increment of y] = [y.sub.t] - [y.sub.t-1]; [a.sub.0] is a 4 x 1 vector of intercept terms; [[phi].sub.i] are the coefficients on the lagged differenced variables contained in [y.sub.t], with p being the optimal number of lags; H is the matrix of cointegrating vectors; [PSI] is the matrix of coefficients on the exogenous dummy variable [w.sub.t], which is equal to 1 when an economic expansion is observed and 0 otherwise; [[mu].sub.t] is a 4 x 1 vector of error terms assumed to be white noise.

Disposable income, corporate profits, and total spending are well-known aggregates and require no further explanation. U.S. disposable income, corporate profits, and total spending (seasonally adjusted, annual rates) were obtained from the U.S. Bureau of Economic Analysis. Revolving credit was obtained from the Federal Reserve which defines it as the sum of revolving credit owned by commercial banks, finance companies, credit unions, savings institutions, non-financial businesses, and securitized consumer revolving credit. Thus this measure of credit is very comprehensive.

All variables are in billions of U.S. current dollars. This model examines the interaction among four key variables in the U.S. economy that played a role in the U.S. economic downturn. The vector autoregression allows us to examine dynamics while the dummy variable captures the asymmetric nature of expansions and contractions of the U.S. business cycles. In effect, this provision makes the VECM model nonlinear. The error correction term is used since the Johansen procedure failed to reject the hypothesis that the variables are cointegrated up to rank2. (2) This shows that a long term (dynamic) relationship exists among the variables over time.

We used quarterly data for the above endogenous variables for the period 1975 to 2007. The large number of observations enabled us to select the optimal lag under the AIC criteria and also allowed us to observe a longer time frame for the dynamic interaction among the variables; a 6 period lag was found to be optimal. Table 1 below shows the Granger causalities among variables.

Taking retail sales as a proxy for aggregate spending, Table 1 indicates that disposable income and revolving credit Granger cause aggregate spending. Spending in turn Granger causes corporate profits. There is, however, some feedback effect, as spending Granger causes disposable income. Revolving credit as well as corporate profits also Granger causes disposable income. All these results are consistent with Keynesian macroeconomics, except that they now highlight the special role of credit in the circular flow of income. The following directed graph (Fig. 1) shows the above causal relationships between variables over time, while Fig. 2 summarizes (in a static format) the probabilistic causalities. Table 2 summarizes the important statistics for the VECM model presented above.

Next, we obtained Generalized Impulse response functions for the above VECM model using Microfit 4.1, developed by Pesaran and Pesaran (1997). With Generalized Impulse response, the ordering of the variables does not matter as opposed to orthogonalized impulse response functions in a VAR model (Pesaran and Shin 1998). We show the details of the Generalized Impulse responses for up to 12 periods in Appendix A.

The Generalized Impulse functions in Appendix A show how negative credit shocks affect spending, disposable income and profits. The VECM model produced an initial one standard deviation shock of $3.98bn in revolving credit. In the actual downturn in 2007, credit declined by about $17bn in the first quarter (or over four times the initial shock obtained from the VECM model) and on average by $5.57bn until 3rd quarter of 2009 (see Table 3).

The VECM model shows an initial shock in spending of $6.07bn, compared to the size of the actual (initial) shock ($5.83bn) being quite dose. However, on average, over eight periods from 4th quarter 2007 to 3rd quarter 2009, actual spending declined on average by $11.64bn, while the largest shock in actual spending for the same time period was $76.33 bn. The largest shock experienced in revolving credit was $23.08bn and $331bn for corporate profits both of which exceed the model's estimated initial shocks. Table 2 thus highlights the actual magnitude of the business cycle downturn in 2007, now labeled "The Great Recession." There was lack of information on the quality of some of the debt, namely securitized mortgages. It was the sudden withdrawal of credit that paralyzed all commerce and led to the greatest downturn since the Great Depression. The sudden withdrawal indicates informational asymmetries in an imperfect financial market governed by fear. Figure 2 also shows that revolving credit was exogenous to the circular flow of income. Revolving credit is determined by outside factors, such as monetary policy, and regulations governing financial institutions that made high leverage ratios possible. In the next section, we provide some important robust tests on our model. First, we examine the effect of choosing a different rank of cointegrating vector on the Granger causal relationships; then we compare our more complex model with other parsimonious models based on SBC and HQC, and finally we examine the Granger causal relationships over different lags.

Testing the Robustness of the Model

In the error correction model presented earlier, the rank of cointegrating vectors was 2. Based on the Johansen procedure at 5 % level of significance, under the trace test, we cannot reject the null hypothesis that the rank of cointegrating vectors is less than or equal to 2 against the alternative that the rank is greater than or equal to 3. However, under the Lambda max test, we cannot reject the null hypothesis that the rank of cointegrating vectors is less than or equal to 1 against the alternative of the rank being equal to 2. Hence, we obtained mixed evidence which indicated a rank of either 1 or 2. Further tests revealed that even if we had chosen r=1 as the rank of cointegrating vectors, the Granger causality structure remained the same; the only exception being that revolving credit does not Granger cause disposable income. Both models (r=l and r=2) indicated white noise residuals and captured some causalities which yielded an augmented version of the circular flow of income. Both r= 1 and r=2 models are useful and most causalities are robust with respect to rank of cointegration. However, the r=2 model does outperform the r= 1 model in most model fit statistics and was chosen based on that.



Next, we examined how the model behaves using a different model selection criterion. One of the principles which guides model building is parsimony. However, we found that for this particular case simple models (lag structure based on BIC and to some extent the HQC criterion) performed worse than more complex model (lag structure based on AIC). The AIC model outperformed the SBC & HQC models in many model fit/selection criteria even though they were simpler. The AIC model captured the dynamics inherent in each variable when we examined actual vs. fitted plots. Also, the AIC model, due to the presence of a large number of lagged dependent variables, automatically corrects for some serial correlation in the residuals, a feature which is lacking in the simpler SBC and HQC models, which had a few equations exhibiting serial correlation.

Finally, we examined the Granger causality structures over different lags. Our purpose here is simply to identify the common (Granger) causal relationships and not an optimal lag. We obtained the Granger causal structures from lags 2 to 8, and from an overall analysis we can point out some common causal structures. Feedback between YD and Spending, Profits Granger causing YD, Revolving Credit Granger causing Sales, and Sales Granger causing profits were the most common causalities observed. There is therefore some degree of robustness with respect to lag structure for the Granger causal relationships. Note, however, that we chose lag=6 based on a specific statistical model selection criteria, the AIC. We avoided an arbitrary lag selection but it can be shown that the AIC performed better than any other selection criteria (as noted above). We also provide a brief examination of the exogeneity of revolving credit in Appendix C. In our Error Correction Model, we found that revolving credit Granger caused disposable income and spending but it was not Granger caused by any other variable. This led us to believe that revolving credit was perhaps exogenous and hence tested this hypothesis in Appendix C. The role of finance is examined in the next section.

Role of Finance and Credit in 2007/09 Contraction

The emergence of a wave of new debt instruments over time can be credited to the dismantling of regulations which began in the 1980s. These controls had been placed on the financial sector in the aftermath of the crash of 1929. In particular, the Depository Institution Deregulation and Monetary Control act of 1980 and the Tax Reform act of 1986 allowed for the emergence of sub-prime mortgages and securitized pools. Assets which are securitized are generally those which can generate steady flows of income over time. For instance, securitized mortgages are called mortgage backed securities (MBS), while assets which are non-mortgage loans but still provide a steady stream of income (e.g., credit card receivables, auto loans, student loans, and lease payments) can be securitized to form asset backed commercial paper (ABCP). Securitization made it possible for firms and financial institutions to immediately realize the value of the income generating asset instead of having to wait long periods of time to access the full amount of the debt. The first set of ABCP originated in 1985 by the Sperry Lease Finance Corporation, which backed its computer equipment leases. (3) Prior to 1980, securitized pools were virtually nonexistent (see Fig. 3) as a form of credit, but they have grown considerably since the inception of ABCP. Commercial banks gradually lost their position as being the major suppliers of credit as the amount of credit supplied in the economy originating from finance companies grew as a result of the deregulation (see Fig. 3).

Dore and Singh (2009) used the Bai-Perron procedure to obtain the best five endogenous structural breaks in: (1) total debt, (2) sum of mortgage and total consumer credit outstanding, and (3) the annual average credit outstanding from 1943 to 2007 (see Table 4). The results show the year in which innovations in the financial market took place. Indeed, the timing of the structural breaks coincided with major legislative acts after allowing for delay or lagged effect, which is common with policy implementation.

Dore and Singh (2009) cited the following acts as the major ones which contributed to the expansion of credit via new debt instruments and other financial innovations:

The Depository Institution Deregulation and Monetary Control Act of 1980 This act phased out most provisions of Regulation Q, which was put in place under the Glass Steagall act of 1933. Regulation Q had placed a limit on interest rates banks could charge and allowed the Federal Reserve to regulate interest rates for savings accounts. This was phased out. The Depository Institution Deregulation and Monetary Control act also allowed for the merger of financial institutions and allowed institutions to charge any interest rates of their choice.

The Tax Reform Act of 1986 This act initiated new low income housing tax credit and allowed interest deductions on mortgage debt but eliminated interest deduction on consumer and auto loans. This encouraged the innovation of instruments allowing lenders to deliver risk adjusted pricing mechanisms instead of having to deny loans to high-risk borrowers. Hence, this act made it easier for innovations in sub-prime mortgages to take place.

The Gramm-Leach-Bliley Act of 1999 (GLBA) This act repealed parts of the Glass-Steagall act, which originally had allowed the Federal reserve to regulate interest rates on savings accounts and prevented commercial banks from trading securities. The Gramm Leach Bliley act repealed parts of the Bank Holding Company act of 1956, which prohibited the merger of commercial banking and insurance institutions. The GLBA allowed commercial entities and investments groups to merge and as a result, institutions provided both banking and insurance underwriting services under one name, e.g., Citigroup. In the process, broad banking developed. Many banking centres offered commercial banking as well as trading securities, investment, and insurance activities--conditions which only occurred in the market prior to the Great Depression. The Glass Steagall act and Bank holding Company act severely limited the ability of banks to enter into insurance and securities markets directly or through subsidiaries (Barth et al. 2000).


The above three acts possibly accounted for the structural shifts in the data generating processes for total debt, (sum of) mortgage, and total consumer credit outstanding and annual average credit for the years 1983, 1988, and 1999. These acts gave birth to other (endogenous) innovations in the financial markets after 1980. In 1987, a new product called Collateralized Debt Obligation (CDO) was issued. CDOs are a type of asset-backed security. CDOs emerged as the fastest growing sector of the asset-backed securities market and became popular with asset managers and investors, which now include insurance companies, mutual fund companies, unit trusts, investment trusts, commercial banks, investment banks, pension fund managers, and private banking organizations. In a low interest rate regime, bonds lost their traditional appeal and the market for derivatives grew dramatically, although derivatives had been around since the 1970s. Derivatives offered investors the incentive to speculate on the movement of the value of the underlying asset over time. These financial instruments tended to be highly leveraged; hence small movements in the in the assets value tended to magnify the speculated value, magnifying the value of the derivative. Hedge funds grew in popularity as a result of deregulation, which allowed innovations to swaps, options, over-the-counter futures contracts, and derivative products. A highly liquid market which occurred as a result of Federal Reserve moving from inflation targeting to interest rate targeting in 1995 provided the environment for the growth and sustenance of these new financial instruments. The Federal Reserve increased the money supply in order to maintain liquidity in the financial markets time and time again. For example, the Federal Reserve reduced its Federal Funds Rate from 6 % in January 2001 to 1% in June 2003 and the Federal Discount Rate from 5.75 % to 2.00 % over the same period. Low interest rates meant that high risk high yielding assets continued to thrive until the downturn. Even during the downturn, the Federal Reserve maintained its stance on interest rate targeting; the Federal Funds Rate was reduced from 5.25 % in June 2006 to 0.25 % in December 2008 and the Federal Discount Rate from 6.25 % to 0.50 % for the same period.

A highly liquid market and the innovation of high-yield, high-risk instruments in the financial market increased profitability in the financial sector at the start of the decade. Coupled with the Economic Growth and Tax Relief Reconciliation act of 2001 (EGTRRA) and The Jobs and Growth Tax Relief Reconciliation act of 2003 (JGTRRA) which reduced corporate taxes and personal income taxes for the uppermost quintile, unprecedented corporate profits were reaped in the U.S. economy prior to the downturn. Figure 4 highlights the movement of both house prices (Case-Shiller Index used as proxy) and corporate profits after tax with the best two structural breaks based on AR1MA models (Dore and Singh 2009). The structural breaks in corporate profits occur in 1993 due to tax decreases and again in 2005. However, the 2005 structural break reflects signs of an impending recession, as profits fell soon after. The structural break in the S&P/Case-Shiller Index of house prices in the U.S. in 2006 reflects the downturn in profits and a weakening economy (see Fig. 4 and Appendix B). For details of these structural break tests, see Appendix B.

One could ask how could there be an economic downturn under the Bush Administration, when there was such favorable treatment of corporate profits, capital gains, personal income, and large government budget deficits, two major fiscal acts designed to stimulate the economy and an expansion in credit. In other words, why did the 2001-2007 expansion come to a sudden end?

Part of the answer lies in the inability to maintain effective demand due to a skewed income distribution. Singh (2008) has shown that for 80 % of U.S. households, income share had declined or remained stagnant in real terms since the start of the decade. Furthermore, current expenditure exceeded current income for the bottom four quintiles since the 1990s, and since most of the population had not seen an increase in their incomes, spending was sustained with credit. Two measures which show the effect of credit on the economy: the Annual Financial Obligations Ratio and the ratio of household debt to GDP have both risen by approximately 5 % since 1980 (Dore and Singh 2009). Non-revolving credit and revolving credit have also grown significantly since 1993. Revolving credit in particular sustained a large portion of the effective demand since consumers were able to borrow repeatedly even in cases where they didn't need to. As a result, over 50 % of sales can be attributed to credit and only 2 % due to disposable income. Hence, when credit dried up aggregate spending also declined sharply. Upon realizing that much of the securitized debt was of poor quality, financial market valuations around the world began to fall. With rising risk premia, banks were hesitant to provide credit for investment and even for meeting payroll purposes. This curtailment of credit was unexpected and large in magnitude; the actual decline in credit was over five times the one standard deviation shock produced from the VECM. With the curtailment of credit, commerce came to a grinding halt- the recession had begun and because the exposure to securitized debt was global in character, the recession was also global in scope. It affected all the major countries whose capital markets were well integrated with the New York capital market.



We began with a purely statistical model, and no a priori economic assumptions were made. We also demonstrate that institutional change can be analyzed in a quantitative way. We did this by showing structural breaks that reflected deregulation of the financial markets. Our VECM model clearly established the causal dynamic and long term relationships among some key macroeconomic variables, such as aggregate spending, disposable income, credit, and corporate profits. The novel feature in the VECM was the demonstration of the role of credit in affecting the circular flow of income. Without imposing a theoretical model, the VECM identifies the key variables and their dynamic interactions. We showed that aggregate spending depends on disposable income and that the latter also depends on spending. Next, it was shown that credit increased disposable income as well as spending. Aggregate spending also raises profits, which in turn raises disposable income. These dynamic interactions provide new hypotheses about the dynamic interactions of macroeconomic variables. Some of these dynamic interactions bear an affinity to Keynesian macrodynamics. However, the important element here is the generation of new hypotheses about macrodynamics. It is these new hypotheses presented here that need to be verified or refuted. In this way, one can determine if the dynamic interactions found here can also be found in the data of other developed countries. In particular, it is important to determine how general is the crucial role of credit, demonstrated in this VECM, in maintaining aggregate spending. Finally, we showed that the decline in credit played a decisive role in the financial downturn and the ensuing Great Recession of 2007/09.

Appendix A
Table 5 Generalized impulse response functions for estimated VECM
(Current US $bn)

Horizon          Sales        YD          Revolving Credit   Profits

Response of variables to a one std error shock in spending:

0                 6.0671        2.598      0.092065            0.66477
1                 5.1301        0.55157   -0.27507             2.9839
2                 5.0996        3.8607     0.022305            7.5466
3                 6.0364        2.1539     0.67399             1.2476
4                 5.6639        9.5206     0.98538            -0.58188
5                 6.4097       14.2961     1.2506             -3.4200
6                 6.0603        8.4737     1.6651             -1.0175
7                 4.8325        8.0375     1.8433              0.73331
8                 4.2402        3.7153     1.9421             -4.2651
9                 3.8798        3.1121     1.6843             -4.6002
10                3.5673        2.6251     1.2204             -2.6242
11                2.9964        0.71097    0.85774            -0.18772
12                2.4316       -1.8237     0.40184             2.5564

Response of variables to a one std error shock in revolving credit:

0                 0.14022      -3.5386     3.9836             -1.0411
1                -0.64496       0.24857    4.7732              0.32175
2                -1.3812       -9.7773     6.2276              6.2478
3                -0.74384     -11.137      7.3147              2.0327
4                -1.756       -15.1279     7.7856              1.4056
5                -2.6976       -9.2175     8.269               1.7412
6                -1.8959      -12.3422     8.7275             -0.81858
7                -2.7825       -9.1799     9.1403              0.61167
8                -2.7885      -10.0357     9.4404             -0.79232
9                -2.8314      -11.9697     9.8373             -1.5826
10               -2.8848      -11.0384     9.9967             -1.2346
11               -2.5923      -10.4614    10.2105             -3.1651
12               -2.2106      -10.0343    10.3978             -3.3265

Response of variables to a one std error shock in YD:

0                 0.37487      42.0479    -0.33525            -6.8835
1                 2.125        26.4893    -0.087628           -5.8875
2                 1.2513       35.0662     0.1876             -3.1344
3                 0.081133     26.8068     0.89362            -8.9836
4                -0.51409      27.6366     1.133             -10.2641
5                 0.9784       25.3453     0.6555            -16.3295
6                 0.84493      28.3116     0.70229           -17.5825
7                 1.0678       26.2017     0.44202           -14.2523
8                 0.86102      23.9954     0.5529            -15.8477
9                 1.15         25.6207     0.62609           -19.236
10                1.5874       26.517      0.58764           -19.7582
11                1.8833       27.3967     0.57031           -18.36
12                2.0017       27.1292     0.69014           -17.5791

Response of variables to a one std error shock in profits:

0                 0.11292      -8.1037    -0.11612            35.7166
1                -0.92854     -10.9818    -0.81435            38.5644
2                -0.01277      -9.6867    -1.0651             25.4431
3                -0.060889      0.10312   -1.4436             35.6206
4                 0.1113        6.5795    -1.7041             45.3343
5                 0.15645       9.1843    -1.2121             46.5462
6                -0.3329        6.0699    -0.93221            47.0269
7                -0.58363       7.9246    -0.71374            44.0943
8                -0.37075       9.5103    -0.43291            43.6942
9                -0.27838      10.0268    -0.3337             45.0145
10               -0.098835      9.0631    -0.13337            41.1139
11               -0.10087       8.3815     0.23522            37.6697
12               -0.0026104     8.3879     0.4971             35.4325

Appendix B

Structural break testing for corporate profits & Case-Shiller Home Price Index
Table 6 Best 2 break points                        Best 2 break points
found using Bai-Perron: ARIMA
(3, 1, 2) for corporate profits
                                  Year             1993:01
                                  BIC              8.164
                                  Sum of Squared   25102.40
Dore and Singh (2009)               Residuals

Table 7 Best 2 break points                        Best 2 break points
found using Bai-Perron: ARIMA
(3, 2, 0) for Case-Shiller Home
Price Index                       Year             1991:02
                                  BIC              0.880
                                  Sum of Squared   143.196
Dore and Singh (2009)               Residuals

Appendix C

In this Section, we examine the effect of future income on contemporaneous revolving credit as well as the extent to which revolving credit is exogenous. In our analysis, the VECM shows the variables which are endogenous, namely the circular flow of income variables. However, revolving credit affects two of them, but it is itself not affected by the other endogenous variables. This suggests that revolving credit is perhaps exogenous. We therefore test the hypothesis that revolving credit is indeed exogenous. We keep the same lag structure (i.e., five lagged periods in first differences) for revolving credit as in our original model (reported in the paper) but we now include them as part of the deterministic or exogenous variables from the beginning of the VECM procedure. Our Johansen cointegration test point to a cointegrating rank of 2. We then utilize the Johansen specification on the covariance matrix for long run structural modelling. We consider this to be a purely statistical approach to the identification problem. As expected, in terms of causal directions, the results from this model from Granger causality testing are the same as in the original model. Table 8 shows the results of the Granger causality tests and the associated p-values. Table 9 shows the block exclusion test for revolving credit as an exogenous group.

Table 9 confirms our original model results. As a group, revolving credit is significant in the spending and income equations as an exogenous component. Similar results (not shown here) were found when only contemporaneous revolving credit (i.e., revolving credit without any lags) was included as an exogenous variable; the exception in this instance was that revolving credit was not significant in the sales equation but all other causal directions remained the same.

Next we test the extent to which agents are forward-looking. This is a purely statistical test and we are specifically examining the effect of future income on current revolving credit. If all agents were forward-looking, then income recipients would expect future profits and all credit suppliers and demanders would expect future profits and hence future (net) income. Using a variation of the Geweke et al. (1982) approach, we test this proposition by including leads of income (YD) for up to six periods (i.e., agents are as forward looking as they are backward looking) in each of the VECM equations as an exogenous component. This would indicate some measure of agent's expectations; in this case the actual future income instead of an equation which makes assumptions on future income expectations. For instance, for the revolving credit equation this would mean that current revolving credit (dependent variable) is dependent on future disposable income (independent variable) as well as the other endogenous variables given above. Our results in Table 10 indicate that the sales, disposable income, and profits equations (circular flow variables) show forward-looking expectations. However, the revolving credit equation does not show dependence on future disposable income. This once again demonstrates that revolving credit is exogenous and not influenced by expectations on future income. Note here that we only used the leads for one variable (YD) and not the leads for every variable in our original VECM. If all of the leads for all of the variables are included and their coefficients are non-zero, then the original VECM would not be a proper specification and the Granger causality results would be compromised. We only isolated future disposable income in order to gain some insight into how it impacts contemporaneous circular flow variables. Including leads in all of the variables will over parameterize our model and we do not have sufficient data to accommodate this. Therefore, our results from the original VECM still hold.

Our overall conclusion from this section suggests that the exogeneity of revolving credit is fairly robust, that it affects the circular flow variables in a unidirectional manner without itself being determined by them. Even if we augment our original VECM model with income expectations, the revolving credit equation should be exempt from this expectation formation. This result in turn reinforces our main observation: the importance of the level of revolving credit in determining the circular flow, and that any curtailment of revolving credit would have a serious impact on spending and hence on the circular flow. This is exactly what happened in 2007-8: when credit was abruptly curtailed and spending and income plunged. The decline was worse than that projected by the generalized impulse response functions, which can only reflect average past response impacts.
Table 8 Granger causality up to 10 % level of significance

                           Variable in column A Granger causes
                                     (with p-value)

Column A            Granger causes   Granger causes   Granger causes
                    corporate        aggregate        disposable
                    profits?         spending?        income?

Corporate Profits   NA               No               Yes
                                     0.181            0.082 *

Spending            Yes              NA               Yes
                    0.008 ***                         0.043 **

Disposable Income   No               Yes              NA
                    0.336            0.000 ***

* p <.10 ** p [less than or equal to] .O5 *** p <.O1

Table 9 Block exclusion test

REVCRED as exogenous group   Wald statistic

Sales equation               CHSQ(5) = 9.3871[.095]
YD equation                  CHSQ(5) = 9.9785[.076]
Profits equation             CHSQ(5) = 6.5407[.257]

Table 10 Block exclusion test

Future YD as exogenous group   Wald statistic

Sales equation                 CHSQ(6) = 19.9943[.003]
YD equation                    CHSQ(6) = 50.3810[.000]
Revolving credit equation      CHSQ(6) = 9.0045[.173]
Profits equation               CHSQ(6) = 15.9117[.014]


Barrell, R., Hurst, I., & Kirby, S. (2008). Financial crises, regulation and growth. National Institute Economic Reviews; 206(1), 56-65.

Barth, J. R., Brumbaugh Jr., R. D. and Wilcox, J. A. (2000). The repeal of Glass-Steagall and the advent of broad banking. Economic and Policy Analysis Working Paper 2000-5.

Bordo, M. D. (2008). An Historical Perspective on the Crisis of 2007-2008. NBER Working Paper, No. w14569.

Cecchetti, S. G. (2008). Crisis and Responses: the Federal Reserve and the Financial Crisis of 2007-2008. NBER Working Paper, No. w14134.

Campos, J., Ericsson, N. R., & Hendry, D. F. (2005). General-to-specific Modeling: An Overview and Selected Bibliography. Board of Governors of the Federal Reserve System, International Finance Discussion Papers Number 838. August 2005.

Davidson, J. E. H., Hendry, D. F., Srba, F., & Yeo, S. (1978). Econometric modelling of the aggregate time-series relationship between consumers' expenditure and income in the United Kingdom. The Economic Journal. 88(352), 661-692.

Dore, M. H. I., & Singh, R. G. (2009). Turning points in nonlinear business cycle theories, financial crisis and the global 2007 2008 downturn. Nonlinear Dynamics, Psychology, and Life Sciences, 13(4), 423-444.

Gorton, G. B. (2008). The Panic of 2007. NBER Working Paper, No. w14358.

Geanakoplos, J. (2009). The leverage cycle. Cowles Foundation Discussion Paper No. 1715. July 2009. Retrieved from:

Geweke, J., Meese, R., & Dent, W. (1982). Comparing alternative tests of causality in temporal systems. Journal of Econometrics, 21, 161-194.

Haavelmo, T. (1944). The probability approach in econometrics. Econometrica, 12, Supplement, iii-115.

Hendry, D. F., & Richard, J.-F. (1983). The econometric analysis of economic time series, International Statistical Review, 51(2), 111-148 (with discussion).

Hendry, D. F. (1987). Econometric methodology: A personal perspective, Chapter 10. In T. F. Bewley (Ed.), Advances in econometrics: Fifth worm congress, Volume 2 (pp. 29-48). Cambridge: Cambridge University Press.

International Monetary Fund (2009). Worm economic outlook: Crisis and recovery, April. See: http:// Retrieved April 26, 2010.

Juselius, K. (2006). The cointegrated VAR model. Oxford: Oxford University Press.

Leamer, E. E. (2007). Housing is the business cycle. NBER Working Paper, No. w13428.

Pesaran, M. H., & Pesaran, B. (1997). Working with Microfit 4.0, interactive econometric analysis. Oxford: Oxford University Press.

Pesaran, M. H., & Shin, Y. (1998). Generalised impulse response analysis in linear multivariate models. Economics Letters, 58, 17-29.

Reinhart, C. M., and Rogoff, K. S. (2008). This time is different: A panoramic view of eight centuries of financial crises. NBER Working Paper, No. w13882.

Singh, R. (2008). United States fiscal policy and its impacts on income distribution. [Mimeo]. Honours Thesis. St. Catharines, Ontario: Brock University.

Taylor, J. B. (2009). The financial crisis and the policy responses: an empirical analysis of what went wrong. NBER Working Paper, No. w14631.

U.S. Bureau of Economic Analysis (2012). National Economic Accounts. Retrieved on May 6, 2012 from: http ://

U.S. Federal Reserve (2012). Federal Reserve Statistical Release, G19, Consumer Credit. Retrieved on May 6, 2012 from:

Woodford, M. (2003). Interest and prices: Foundations of a theory of monetary policy. Princeton: Princeton University Press.

Woodford, M. (2009). Convergence in macroeconomics: Elements of the new synthesis. American Economic Journal: Macroeconomics, 1(1), 267-279.

M. H. I. Dore ([mail]) * R. G. Singh

Department of Economics, Brock University, St Catharines, ON, Canada L2S 3A1


(1) See Economic Growth and Tax Relief Reconciliation Act of 2001 (EGTRRA) and The Jobs and Growth Tax Relief Reconciliation Act of 2003 (JGTRRA)

(2) Under trace test. Lambda max test failed to reject hypothesis that rank = 1. Further tests (not shown) reveal r = 2 has a better model fit compared to r=1.

(3) According to Statement of Cameron L. Cowan before tile Subcommittee on Housing and Community Opportunity Subcommittee on Financial Institutions and Consumer Credit. United States House of Representatives. 2003.

This research is in part funded by the Social Sciences and Humanities Research Council of Canada. An early draft of this paper was presented at the 8th OxMetrics User Conference, The George Washington University, Washington, DC, March 18-19, 2010. This revised version was presented at the 73rd International Atlantic Economic Society Conference held at Bahceschir University, Istanbul, Turkey, 28-31 March, 2012. The authors are grateful for comments received at the conferences, but they alone are responsible for its contents.

Published online: 23 June 2012

DOI 10.1007/s11293-012-9326-2
Table 1 Granger causality up to 10 % level of significance

                       Variable in column A Granger causes
                                  (with p-value)

Column A            Granger causes       Granger causes
                    corporate profits?   aggregate spending?

Corporate profits   NA                   No

Spending            Yes                  NA
                    0.011 **

Disposable income   No                   Yes
                    0.316                0.000 ***

Revolving credit    No                   Yes
                    0.236                0.009 ***

                      Variable in column A Granger causes
                                (with p-value)

Column A            Granger causes       Granger causes
                    disposable income?   revolving credit?

Corporate profits   Yes                  No
                    0.011 **             0.486

Spending            Yes                  No
                    0.045 **             0.109

Disposable income   NA                   No

Revolving credit    Yes                  NA
                    0.084 *

Table 2 Model fit statistics for VECM model (r=2)

                                  Sales             YD

R-Squared                         0.48963           0.48578
R-Bar-Squared                     0.37455           0.36983
S.E. of Regression                6.0073            41.6336
Mean of Dependent Variable        7.8135            71.9603
Residual Sum of Squares           3680.9            176802.2
F-stat. F(23, 102)                4.2546[.000]      4.1895[.000]
Equation Log-likelihood           -391.3887         -635.3161
LM test for serial correlation:
CHSQ (1) with p-value             .3234E-3 [.986]   0067956[.934]

                                  Revered        Profits

R-Squared                         0.627          0.46244
R-Bar-Squared                     0.54289        0.34122
S.E. of Regression                3.9444         35.3647
Mean of Dependent Variable        7.3367         13.1397
Residual Sum of Squares           1586.9         127567.5
F-stat. F(23, 102)                7.4546[.000]   3.8150[.000]
Equation Log-likelihood           -338.3826      -614.7538
LM test for serial correlation:
CHSQ (1) with p-value             .97533[.323]   .66667[.414]

For the LM test, the null hypothesis is zero first order serial
correlation in the residuals

Table 3 Actual vs estimated shocks from the VECM model (Current
US $bn)

(in bn of current dollars)     Spending     Disposable

Estimated initial one S.D.     -6.07        -42.05
Shock from VECM model

Size of actual initial Shock   -5.83        -117.40
(after 4th Quarter, 2007)

Average quarterly Shock from   -11.64       -52.42
2007 Q4 to 2009 Q3

Largest Actual Shock between   -76.33       -117.40
2007 Q4 to 2009 Q3

Timing of largest Shock        2008 Q3-Q4   2008 Q3-Q4

(in bn of current dollars)     Revolving         Corporate
                               credit            profits

Estimated initial one S.D.     -3.98             -35.72
Shock from VECM model

Size of actual initial Shock   -17.82            -39.70
(after 4th Quarter, 2007)

Average quarterly Shock from   -5.57             -17.56
2007 Q4 to 2009 Q3

Largest Actual Shock between   -23.08            -331.00
2007 Q4 to 2009 Q3

Timing of largest Shock        2008 Q4-2009 Q1   2008 Q2-Q3

Table 4 Bai-Perron structural break tests from Dore and Singh (2009)

              Total debt     Sum of mortgage   Annual average
                             debt and          credit
                             consumer credit   outstanding

              Best 5 break   Best 5 break      Best 5 break
              points         points            points

ARIMA Model   (3,2,0)        (3,2,0)           (3,2,0)

Year          1983           1983              1983
              1989           1988/89           1987
              1993           1992/93           1991
              1997           1999              1999
              2002/03        2003              2003
COPYRIGHT 2012 Atlantic Economic Society
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2012 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Dore, Mohammed H.I.; Singh, Rajiv G.
Publication:Atlantic Economic Journal
Geographic Code:1USA
Date:Sep 1, 2012
Previous Article:Mortgage product substitution and state anti-predatory lending laws: better loans and better borrowers?
Next Article:Explaining the passage of living wage legislation in the U.S.

Terms of use | Privacy policy | Copyright © 2022 Farlex, Inc. | Feedback | For webmasters |