direct empirical evidence that QuickPay is good

[JNing0]

So far we have come up with theories with the over-arching theme that the project delay we observe is an efficient response to QuickPay. So QuickPay helps. But this positive spin exists only in our theories. To strengthen our case, can we find direct evidence in our data that proves this? One way is to look at variables other than project delay. As liquidity constraints play an important role in our theories, a natural candidate is default rate. Our earlier attempts in this failed. Are there other avenues to get to it? What are other things we can consider based on our theories?

[vob2]

Good question. Although one quibble is that an efficient response does not mean the outcome is good. Equilibrium does not have to be value enhancing. The response from the supplier could be efficient and it still hurts the government. In other words, I do not think in our theories we postulate that QuickPay is good or bad. We just try to explain how it would affect the delay (or another other metric). Could you please clarify what you mean by "positive spin"?

I had a question about defaults (Issue #28). But it looks like most companies are private and we have little overlap with public data, such as COMPUSTAT and data on defaults (LoPucky default database). There is a field in USASpending that marks contract termination cause, and one option is for Default. But only 0.1 or 0.01 of contracts have that marked.

The question is the right question though. Empirically QuickPay seems bad for the government. Are there any redeeming features that justify this program?

[JNing0]

You are right. The notion "QuickPay is good" is meaningless.

Two parties are directly affected by QuickPay: the supplier and the government. The supplier's other customers and the supplier's upstream suppliers are indirectly affected by QuickPay. The supplier is profit-maximizing. The objective of the government is more complicated. It should include the timeliness of the project, the quality of the project, and the "well-being" of the economy, which is the motivation for QuickPay and includes its indirect influences.

All theories establish that a liquidity-constrained supplier should delay under QuickPay to maximize profit. So QuickPay is good for the supplier.

QuickPay hurts the government in project timeliness. But it may increase project quality, helps with the supplier's liquidity constraints, and helps the supplier to grow its other businesses. These positive outcomes of QuickPay for the government is what I meant by "positive spin."

Which of these positive effects can we test? Quality seems difficult. Liquidity constraints could be tested by default rate, which does not seem easy based on earlier discussions. We may use alternative metrics to measure a firm's financial constraints, e.g., sales? What about the entry of small businesses to government projects? A 30-day payment delay implies a high entry barrier on firm's working capital whereas a 15-day delay reduces the barrier? The number of projects that receive contract financing?

[harishk05]

Thanks for these notes. When writing up my theory note, I did consider adding a government payoff function (which would be different that the contractor payoff function, of course, and would include other features like timeliness -- see Bajari and Lewis paper in the literature -- and quality of the project). This would have allowed us to model the "socially optimal" project completion time (or at least the optimal project completion time for the government/principal).

This would then become a principal-agent or mechanism design model, which would lead to questions about what the optimal contract that the government should offer the contractors.

It is not clear to me that we want to take this paper in that direction. Of course, this is a natural question and one we should discuss.

Here is one idea.

I need to think more about this, but I wonder if we can separate the effect of Quickpay and contract financing. What I mean is the following.

In the note I uploaded, the new Theory 3 has a parameter \alpha. Roughly speaking, this parameter represents the ability of the firm to use other forms of financing in order to meet the liability that is due at time t_L. (If \alpha = 1, there is no ability to use other forms of financing and the liability must be satisfied exactly at time t_L. If \alpha > 1, then the contractor can push out this payment.) In other words, contract financing provided by the government (or other sources) increases \alpha.

On the other hand, Quickpay reduces \tau. (Note that t_1 + \tau <= \alpha*t_L.)

Suppose now that the supplier is facing a constrained optimal situation, where the project length is constrained by the liability payment time. Note that the liability payment time is \alpha*t_L.

In other words, t_1^C + \tau = \alpha*t_L.

The government can try to "help" the contractor by reducing \tau (Quickpay) or increasing \alpha (contract financing). If the government does not offer contract financing (\alpha is fixed) and offers Quickpay (\tau is reduced) then, as long as the constraint binds, there is a 1:1 substitution between \tau and t_1^C. If the government pays 1 day sooner, the project will be delayed by 1 day.

Instead, suppose the government raises \alpha. Again, as long as the constraint is binding, an increase in \alpha will increase t_1^C.

But suppose \alpha is increased to the point that this constraint is no longer binding. In other words, suppose \alpha is above some threshold and t_1^U + \tau < \alpha*t_L. (Note that t_1^U is the unconstrained optimal project duration for the supplier.)

In this case, reducing \tau will reduce project duration (by the strategic complementarity argument).

Basically, my point is the following. For the government, the effectiveness of Quickpay depends on the financial health of the contractor. For a financially healthy contractor, Quickpay will speed up projects (assume that this is what the government wants). However, for a financially constrained contractor, Quickpay will delay projects. But note that the government has tools to directly influence the financial health of the contractor. The government can offer contract financing (which affects \alpha). Doing so may be prudent in the following sense. If the government can raise \alpha to the point that the liability constraint is never binding, then Quickpay can be used to speed up projects. However, "small" increases in \alpha will actually end up delaying projects. The government has to commit to a large enough increase in \alpha for Quickpay to reduce project duration. In summary, the decision to offer contract financing and Quickpay should be linked.

Not sure how this fits into our empirical analysis. But maybe worth discussing.

[JNing0]

Another thought on the positive effect of QuickPay. Our theories state that it is optimal for a financially constrained supplier to delay government project because doing so improves its profit. We can test this by studying the effect of QuickPay on the sales of a business, i.e., using sales as the response variable. Or even better, directly regress sales with respect to project delays.

Sales feels a better measure of a firm's well-being than whether or not a firm receiving contract financing or financial assistance, because sales is a continuous variable whereas the latter are 0-1 variables. Sales is also better than the number of suppliers who receives financing because the latter requires aggregation across firms, which would leave us with only the time dimension. In contrast, sales is a firm-level variable, so we have a much larger data set and thus more statistical power.

[vob2]

What do Sales measure? Total Sales of the firm or Sales to the government?

Why are you comparing Sales with receives financing indicator? The two serve different purposes, no? I not following you there.

[JNing0]

I meant total sales of a firm. So in the regression, our response variable _Yit would be the total sales of firm i in quarter t. We have this information in our data set. In the regression, we may need to have a lag to see the effect. For example, QuickPay or delaying the project this quarter improves the firm's sales next quarter.

Let _Sit denote the total sales of firm i in quarter t and let _RFit denote the indicator of whether firm i receives financing assistance in quarter t. _Sit and _RFit are similar in the sense that they are both measures of a firm's well-being. Intuitively, if a firm is doing better, then its sales increases and it does not need financing as much. So we can test whether QuickPay indeed helps a firm by examining how QuickPay affects _Sit and/or how QuickPay affects _RFit. Testing _Sit is easier than testing _RFit because the latter is binary.

[JNing0]

Another variable that we can look at is "Initial value of contracts." Intuitively, larger businesses have better financial health, larger workforce, more resources, etc. that allow them to take on larger projects with higher values. This intuition is confirmed by the graph here.

If the 2011-QuickPay did help small businesses grow (in financial capability, workforce, etc.), then they should be able to score contracts with higher values (seems to be the case in the graph above).

By definition, Contract's initial value = "base and all options value" on the date when the contract first appeared in the sample. So each contract provides one observation. We can look at a statistic of values of all contracts owned by a firm in a particular quarter.

For example,

• Response variable _Yit = avg. (or median or max.) value of all contracts owned by firm i in period t.
• Treatment group _Ti = 1 if firm i is categorized as small business under at least one of its contracts

We can further divide the treatment group based on the effect of QuickPay on the firm. One way is to look at the number of contracts that are treated under QuickPay. Another better approach is to use the "business reliance" metric we created earlier. We should compute this reliance metric, i.e., the fao-to-sales ratio, using fao from treated projects. For this analysis, we can make it really simple by considering two instead of three groups.

Let _HIit be a dummy variable that equals 0 if treated contracts has low impact (small weight) in firm i's sales in period t and equals 1 otherwise.

We won't have the _HIit variable in firms in the control group as such a firm is always labeled "large business."

So we will use a slightly different model than before

Note that this model no longer has the beta_4 term in here.

The coefficient of interest is beta_2, which tells us the value increase in a firm's projects if it benefits more from QuickPay.

Relatedly, the reliance metric in the business reliance regression should be computed similarly using treated projects. In the earlier version of the regression, we did not need to make the distinction because we looked at firms that receive only one type of contracts. Maybe that's the reason for the big difference between the latest result and the earlier one? I think we should use model (1).

[vob2]

Sounds like a good idea