2.2 Empirical issues on securitization
2.2.5 Difference-in-difference analysis
Apart from the empirical strategies above, another popular estimation method is called the Difference-in-Difference analysis. The simplest set up is one where outcomes are observed for two groups for two periods. One of the groups is exposed to a treatment in the second period but not in the first period. The second group is not exposed to the treatment during either period. In the case where the same units within a group are observed in each time period, the average gain in the second (control) group is substracted from the average gain in the first (treatment) group. This removes biases in second period comparisons between the treatment and control group that could be the result from permanent differences between those groups, as well as biases from comparisons over time in the treatment group that could be the result of trends. To understand the Difference- in-Difference analysis, it would be better to start from the basic fixed-effects model. In the fixed effects models, if a researcher is interested whether πππ is
affected by π«ππ‘ which is assumed to be randomly assigned. There are also time
varying covariates πΏππ‘ and unobserved but fixed confounders π¨π. Therefore,
π¬[ππππ|π¨π, πΏππ‘, π] = πΆ + ππ+ π¨πβ²πΈ + πΏππβ² π·. (2.52) Assuming that the causal effect of individuals is additive and constant so the following equation is also true:
π¬[ππππ|π¨π, πΏππ‘, π] = π¬[ππππ|π¨π, πΏππ‘, π] + π (2.53) Taken together, we will have:
π¬[ππππ|π¨π, πΏππ‘, π] = πΆ + ππ+ ππ«ππ‘+ π¨πβ²πΈ + πΏππβ² π· (2.54) This equation implies the following regression equation:
πππ = πΆπ+ ππ+ ππ«ππ‘+ πΏππβ² π· + πΊππ (2.55)
Where πΊππ= ππππβ π¬[ππππ|π¨π, πΏππ‘, π], and πΆπ = πΆ + π¨πβ²πΈ.
Suppose we simply estimate this model with OLS without fixed effects, then the estimation is:
πππ= ππππππππ + ππ+ ππ«ππ‘+ πΏππβ² π· + πΆπ+ πΊππ (2.56)
As πΆπ is correlated with the individual status π«ππ‘, there is a correlation of
π«ππ‘ with error term. This will lead to biased OLS estimations. A fixed effect model
would address this problem because πΆπ would be included in the regression. π«ππ‘
with error term would therefore be uncorrelated and the regression would obtain
an unbiased estimator π.
In practice, there are two ways of estimating the fixed effects model: i) demeaning, or the within estimator, and ii) first differencing. With demeaning we should first calculate individual averages of the dependent variable and all explanatory variables. The we should substract the averages from the regression to obtain:
πππβ πΜ π= ππβ πΜ + π(π«ππ‘ β π«Μ π) + (πΏππ‘β πΏΜ π)β²π· + πΆπ+ (πΊππβ πΊΜ π) (2.57)
Thus πΆπ drops out and therefore the error and the regressor would no longer
be correlated.
In the first differencing way, we can also get rid of the πΆπ by:
π«πππ = π«ππ + ππ«π«ππ‘+ π«πΏππβ² π· + π«πΊππ (2.58) The Difference-in-Difference method is first introduced by Card and Krueger (1994) who analyse the effect of a minimum wage increase in New Jersey. Taken securitization as an example, we can obtain a bank securitizes loans or not. We can only observe one situation or the other, that is, at a time point, a bank can only be a securitizer or a non-securitizer, but cannot be both.
If we assume that: 1) π1ππ π‘ is the performance indicator of bank π which has
securitized assets at state π and time π, and 2) π0ππ π‘ is the performance indicator
of bank π which does not have securitized assets at state π and time π. We the assume that:
π¬[πππππ|π, π] = πΈπ + ππ (2.59)
In the absence of the securitization activities, a bankβs performance is
determined by the sum of a time-invariant state effect πΈπ and a time effect ππ.
Let π«π π‘ be a dummy for securitized banks after an endogenous shock, e.g., the
bankruptcy of Lehman Brothers in 2008. Assuming π¬[πππππβ πππππ|π, π] = πΉ is the treatment effect, observed bank performance thus can be written as:
ππππ = πΈπ + ππ+ πΉπ«π π‘+ πΊπππ (2.60) For example, for banks with securitization activities, the performance before the bankruptcy of Lehman Brothers in 2008 is:
π¬[ππππ|π = πππππππππππ, π = ππππππ ππππ] = πΈπππππππππππ+ ππππππ (2.61)
And the performance after the bankruptcy of Lehman Brothers in 2008 is:
π¬[ππππ|π = πππππππππππ, π = πππππ ππππ] = πΈπππππππππππ+ πππππππ (2.62)
Therefore, the difference between the securitizersβ performance before
and after 2008 is:
π¬[ππππ|π = πππππππππππ, π = πππππ ππππ] β
π¬[ππππ|π = πππππππππππ, π = ππππππ ππππ] = πππππππβ ππππππ+ πΉ (2.63)
Similarly, for non-securitized banks, the performance before the bankruptcy of Lehman Brothers in 2008 is:
π¬[ππππ|π = ππππππππππππππ, π = ππππππ ππππ] = πΈππππππππππππππ+ ππππππ (2.64) And the performance after the bankruptcy of Lehman Brothers in 2008 is:
π¬[ππππ|π = ππππππππππππππ, π = πππππ ππππ] = πΈππππππππππππππ+ πππππππ (2.65)
Therefore, the difference between the securitizersβ performance before
and after 2008 is:
π¬[ππππ|π = ππππππππππππππ, π = πππππ ππππ] β
π¬[ππππ|π = ππππππππππππππ, π = ππππππ ππππ] = πππππππβ ππππππ (2.66) Finally, the Difference-in-Difference strategy allows us to compare the change in the performance of securitizers with the change in the performance of non-securitizers. The population Difference-in-Difference is:
{π¬[ππππ|π = πππππππππππ, π = πππππ ππππ] β π¬[ππππ|π = πππππππππππ, π = ππππππ ππππ]} β
{π¬[ππππ|π = ππππππππππππππ, π = πππππ ππππ] β
π¬[ππππ|π = ππππππππππππππ, π = ππππππ ππππ]} = πΉ (2.67)
The advantages of the Difference-in-Difference method are stated as follows. First, it is easy to calculate standard errors under this framework. Second, it allows researchers to control for other variables which may reduce the residual variance, which could also lead to smaller standard errors. Third, it is also easy to include multiple periods. Last, researchers can study treatments with different treatment intensity.
A typical regression model that can be estimated under the Difference-in- Difference framework is presented as follows:
πΆππππππππ= π·π+ π·ππ»πππππππππ+ π·ππ·ππππΊπππππ+ π·π(π»ππππππππ Γ π·ππππΊππππ)ππ+ πΊ (2.68)
Where π»ππππππππ is the dummy if a bank confirmed as a securitizer, while