Results for portfolio theory

QuickPay-Operational-Performance / Data-and-code

Data and code for econometric analysis

0 stars 0 forks source link

Results for portfolio theory #86

Open vibhuti6 opened 2 years ago

vibhuti6 commented 2 years ago

Hi everyone, here's a summary of results for the portfolio theory (code here). Please review and comment. Thank you!

I created a variable to measure the proportion of large projects held by the contractor in a given year-quarter:
- Proportion of Large Projects = Num of large projects/Total number of projects
I interacted the treatment effect with this variable, using two specifications:
- on the full sample, and
- on a restricted sample where contractors hold at least one large and one small project in a given year quarter (so sequencing of projects is feasible)
We see some effect with the restricted sample, but it's not very robust.

I also ran the discrete version of the regression (with all controls and fixed effects) based on quintiles which indicates a null effect.

vob2 commented 2 years ago

Question: how do you define a large vs small project? The same way we classify large vs small contractors? So the large project is not subject to expedited payments but small one is?

Not sure why we get significance in the restricted sample but not in the full sample. It looks like there are a lot of observations that are only small or only large projects. Maybe there is not enough variation in the variable proportion of large and proportion of small projects. In a sense, having this variable either 0 or 1 is the base case. So maybe there is a non-linear relationship but we are trying to capture it as linear. As proportion increases from 0 to 0.5, the portfolio effect increases, but if the proportion continues to increase to 1, then the effect decreases.

vob2 commented 2 years ago

Perhaps we should make the model non-linear? Add quadratic term for the proportion of large projects.

JNing0 commented 2 years ago

Nice work, Vibhuti! I know we decided to only keep one version of the portfolio theory but I don't remember which one. So my comment below may be off...

Is there a specific reason for computing the fraction of large projects by enumeration instead of project value, e.g., the initial value? I feel that the latter is a better metric for the effect of QP on a business portfolio and thus the incentive for a contractor to sequence.

vibhuti6 commented 2 years ago

Thank you both!

Not sure if making it non-linear would make much of a difference, as we don't see an effect in the discretized version.
The restricted sample may be more appropriate to test this theory because it excludes contractors holding only small or only large projects. For these contractors, there isn't any payment disparity between projects that would incentivize strategic sequencing.
The results are similar if we use project value as the metric. Please see below.

vob2 commented 2 years ago

I like the results on the restricted sample, obviously, although it does go against the theory I just advocated for another issue.

@vibhuti6 Why are there two rows for TreatPostProportion of the large project budget in the tables? One of them must be Treat*Post. Could you please double-check that the numbers are correct though?

vibhuti6 commented 2 years ago

Hi Vlad, thank you for catching this -- it was a typing error. Please see the updated table below.

vibhuti6 commented 2 years ago

Hi everyone, I ran the regressions based on pre-defined thresholds (as we discussed). There's no effect with this model either -- please see below. Thanks!