A guest post by Mark Sadowski
Atif Mian and Amir Sufi have a post that is provocatively titled “Subprime Lending Drives Spending” and which opens with:
“A concern that we highlighted in yesterday’s post is that the only way the U.S. economy can generate significant consumer spending is through aggressive lending to borrowers with low credit scores. Here is more evidence supporting that view.”
and which concludes with:
“It appears that the key to boosting spending in the U.S. economy is subprime lending. The financial system was lending against homes before the Great Recession, and now it has moved to lending against cars. But the basic message is the same.”
These are assertions about the macroeconomy which Sufi and Mian support with what I would characterize as borderline microeconomic evidence. I have no doubt that there is a strong correlation between mortgage loans and appliance, furniture and home improvement spending, and between auto loans and new auto spending. I also have no doubt that during periods of economic upswing that much of the increase in lending may be subprime. However they provide no evidence in their post that it is in fact subprime lending which is driving home and auto spending, and not home and auto spending that is driving subprime lending.
The reason why I bring this up is that one of the most robust results I have discovered, in my investigations into the relationship between lending and spending on the macroeconomic level is that changes in spending precede changes in lending, and not the other way around. In particular, changes in private nonresidential fixed investment precede changes in business sector credit market debt and bank lending, and changes in private residential fixed investment precede changes in bank lending.
To illustrate this general result let’s look at the history of U.S. household sector lending and spending. Here is nominal household sector credit market debt and nominal personal consumption expenditures and private residential fixed investment at quarterly frequency since 1951Q4 in log levels.
And here’s a scatterplot of the year-on-year percent changes in each along with the elementary ordinary least squares (OLS) results..
The correlation is statistically significant at the 1% significance level. But what we can’t tell simply from looking at this graph or from the OLS results is whether it is lending that is causing spending or if it is spending that is causing lending.
One established way of testing causality (Sims 1972) is to construct a bivariate Vector Auto-Regression (VAR) and test for Granger-causality. The following analysis is performed using a technique developed by Toda and Yamamato (1995).
The data I used is nominal household sector credit market debt (CMDEBT) and nominal personal consumption expenditures and private residential fixed investment (PCEPRFI) in log levels. Using the Augmented Dickey-Fuller (ADF) and Kwiatkowski-Phillips-Schmidt-Shin (KPSS) tests I find that the order of integration for each series is two. I set up a two equation VAR in the levels of the data including an intercept for each equation. Most information criteria suggested a maximum lag length of seven for each variable. An LM test suggests there is no problem with serial correlation at this lag length. Incidentally, an AR roots graph suggests that the VAR is dynamically stable and Johansen’s Trace Test and Maximum Eigenvalue Tests both indicate the two series are not cointegrated.
Then I re-estimated the levels VAR with two extra lags of each variable in each equation. But rather than declare the lag interval for the two endogenous variables to be from 1 to 9, I left the interval at 1 to 7 and declared the extra two lags of each variable to be exogenous variables. Here are the Granger causality test results.
|VAR Granger Causality/Block Exogeneity Wald Tests|
|Date: 06/14/14 Time: 09:57|
|Sample: 1951Q4 2014Q1|
|Included observations: 241|
|Dependent variable: PCEPRFI|
|Dependent variable: CMDEBT|
I fail to reject the null that nominal household credit market debt does not Granger cause nominal personal consumption expenditures and private residential fixed investment, but I reject the null that nominal personal consumption expenditures and private residential fixed investment does not Granger cause nominal household credit market debt at the 1% significance level.
In other words U.S. household sector spending provides statistically significant information about future household sector lending, but not the other way around. As I said, the finding that spending precedes lending at the macroeconomic level is fairly robust across a variety of contexts, so anytime I see anyone jumping to the conclusion that lending causes spending merely based on their correlation, it causes me to cringe.
Do the Granger causality results actually mean that spending “causes” lending? No, in particular it could mean that a third variable is causing both lending and spending. But it does cast serious doubt on the idea that lending always causes spending based merely on their correlation.
On the other hand, there are strong theoretical reasons why we might expect spending to precede lending. At the macroeconomic level everyone’s spending is someone else’s income. The ability and willingness to take on debt arguably is income constrained. Thus we might expect to see changes in income and spending precede changes in lending.
And in fact if it is monetary policy which is the cause of nominal income and spending, as I strongly believe to be the case, then increased lending is not the key to boosting the U.S. economy. Rather increased lending is simply one possible consequence of an adequately expansionary monetary policy.