Q about specific Difference-in-Difference design with unbalanced Panel data

[Max Quitra]

Hi all,
I am currently thinking about an empirical design which im unsure of if it is really a DiD approach. My empirical analysis is build upon loans originated between 2010 and 2015. For each of the loans originated in that time span, I observe the daily pricing data on secondary markets. Now in 12/2013 there was a policy intervention, which changed how loans will be structured in the future. That is, all loans originated after 12/2013 are the treated loans which are affected by the intervention and for which we expect a price reaction.

In a standard DiD design we observe a treatment group and a control group, and for both groups we have data before and after the intervention. However, in my special case, the treated units cannot be observed before the treatment. Is this still a valid DiD in design? Or is it just standard pooled OLS, which has the downside that loans originated after the intervention might be priced differently than the old existing ones, even if there was no intervention?

If it is indeed a properly idendified DiD design, I would really appreciate if someone could link me to some papers that talk about such a design or even implement it on their own.

Best
Max

[Jared Greathouse]

No pre-treatment days=no causal inference. At least in the panel setting, you NEED pre-intervention data for any kind of counterfactual.

[Max Quitra]

Okay, thanks for your decicive answer.

Now, maybe we can go a step further, as I have just described the base scenario. What if some of the loans originated post 12/2013 posess some very specific feature, and if they posess this very specific feature, they are even more affected by the intervention. The idea then is to use "Having the Feature" as Treatment Variable and making the time of intervention just the Pre/Post Variable (even though it affects each loan originated post 12/2013). Because originated loans do not immediately acquire this feature, could identification be build upon variation in treatment timing? So that a causal effect is based upon comparing loans from the treatment group that were already treated to loans from the treatment group that are not yet treated?