RDP 2013-05: Liquidity Shocks and the US Housing Credit Crisis of 2007–2008 6. Testing the Flight to Quality and Flight to Home Hypotheses

6.1 Identification

I now re-estimate the benchmark equation but augment it with indicators of borrower risk to determine how the effects of the liquidity shock vary across different classes of borrowers. I estimate the following panel regression model:

The dependent variable (ΔL_ij) is again a measure of the percentage change in lending activity during the crisis for lender i in tract j. Amongst the independent variables, I again include the liquidity shock variable (SALESHARE_i) but I now also interact this variable with indicators for whether the lending is risky or not and the distance between the borrower's home tract and the location of the lender's headquarters. More specifically, as explanatory variables I include a variable that measures the ratio of subprime lending to total lending (SUBPRIME_ij) of lender i in tract j and a measure of the (log) number of kilometres between the headquarters of lender i and tract j (DISTANCE_ij). I also include the same set of pre-crisis control variables (X_ij) as in the benchmark equation.

If the liquidity shock caused a flight to quality, the coefficient on the interaction variable SUBPRIME * SALESHARE will be less than zero (β₁ < 0) as OTD lenders shift lending away from subprime borrowers by more than non-OTD lenders. If the liquidity shock caused a flight to home, the coefficient on the interaction term DISTANCE * SALESHARE will be less than zero (β₂ < 0) as OTD lenders disproportinately reduce lending to distant borrowers. These hypotheses do not, however, rule out a more general flight to quality or flight to home by all lenders, but simply that these effects are not driven by the liquidity shock.

6.2 Graphical Analysis

Splitting lenders into the two groups – OTD and non-OTD lenders – within each tract and year, I now examine how different types of lending evolved. In Figure 5, I plot the evolution of the share of subprime lending for both OTD and non-OTD lenders. There is some evidence that the OTD lenders reduced their exposure to subprime lending by more than the non-OTD lenders around the time of the crisis. The overall share of subprime lending was broadly similar for the two groups in the pre-crisis period but a wedge emerged in 2007 as the subprime lending of the OTD lenders shrunk while the subprime lending of the non-OTD lenders remained elevated. The difference in the share of subprime lending persisted through the post-crisis period. This shift away from risky lending to less-risky lending is consistent with a flight to quality by the mortgage lenders most affected by the liquidity shock.

In Figure 6, I plot the evolution of the average lending distance for both groups. There is little aggregate evidence that the OTD lenders shifted their mortgage lending towards borrowers that were closer to their head offices by more than the non-OTD lenders. The conditional estimates indicate that, on average, the OTD lenders originate loans at more than twice the distance of non-OTD lenders. The average lending distance of the OTD lenders fell slightly relative to the non-OTD lenders in 2007, but quickly recovered the next year. In other words, based on this simple graphical analysis, there is little evidence in favour of the flight to home effect driven by the liquidity shock.

6.3 Econometric Analysis

The liquidity shock stemming from the securitisation market had a sizeable negative impact on mortgage lending. Estimates of Equation (3) show which borrowers were most affected by the shock (Table 3).

The negative coefficient on the SUBPRIME variable indicates that there was a tendency for all lenders to reduce credit to subprime borrowers during the crisis. Moreover, the negative coefficient on the SUBPRIME * SALESHARE interaction variable suggests that OTD lenders cut lending to subprime borrowers by more than the non-OTD lenders. However, the differential effect is not statistically significant. In other words, there is evidence that all lenders reduced credit to subprime borrowers, rather than just the OTD lenders, which is inconsistent with the specific flight to quality hypothesis considered here. The positive coefficient on the DISTANCE variable suggests, surprisingly, that there was a tendency to increase lending to more distant borrowers during the crisis, although the negative coefficient on the DISTANCE * SALESHARE interaction variable implies that the OTD lenders cut lending to more distant borrowers by more than the non-OTD lenders. However, the differential effect is, again, not statistically significant. In other words, there is limited evidence to support the flight to home hypothesis that the change in lending behaviour is caused by the liquidity shock.