Thank you very much.
All right. Really great to have this opportunity to present some work and engage in discussion. I’ve already benefited from the previous speakers. So the question is how does increase in financial infrastructure affect inequality and shared prosperity? I’m going to give a couple of answers to that direct question: a historical answer with regards to the spread of branches of bank, so called brick and mortar banking; a contemporary answer, having to do with digital banking. These results are based on a paper with Ji and Teng.
A bit about the model, I will say that this talk is quite specific. I take the models quite seriously, which is not to say that they're not leaving things out. But I'm letting the model kind of engage in a dialogue. The model will speak back in terms in this case of these policies.
In terms of this model, the period is Thailand 1986 to 2011. You can identify actual and potential branch locations, namely where banks are or eventually located their branches. There's a lot of local markets to study and banks in the model decide where to put their bank, their branches and when. There's a spatial cost here, which is salient to access the branch, either for credit as firms borrow or for saving among households are or firms saving up for the future. There's also a collateral constraint to the model that distinction matters. And I'll come back to that.
What we do is to calibrate the model based on observed 1986 initial conditions, and then stimulate and compare to what actually happened in the data. We're going to get very close to the data so in some sense, the model is validated. Then we can use it to run this digital banking experiment as a counterfactual where we're going to drop those spatial access costs to zero. The importance of this kind of work, not just for me, is you get lessons from the data and the model. Although I put it on this slide in the context of the first model, it's quite a general point. As I go through the subsequent slides, you can identify the dynamics, the transition paths, the delays, the uneven effects across local markets.
And then we come to this terminological, and definitional issue of shared prosperity. Inequality, I would say, is not necessarily a bad metric, but it can be misunderstood if it's not put in the context of some people can be better off, but more unequal. So inequality goes up while shared prosperity is going up. And I will say that in this model, there's a lot of “sharing” that happens through substantial intergenerational flow of funds. So there is advice to policymakers here, and in the subsequent slides, which is to be clear about these distinctions. In particular, financial access is not a panacea, and policymakers are going to have to build in some commitment against political pressure, if possible, or choose another path. And the model based approached delineates those options. The model is successful in the sense of generating cumulative growth of about 10% per year. That's attributable to 41% growth in labor and 24% growth in capital. The model is featuring occupation transitions, as is typical in developing economies from farming to wage work or to setting up or expanding small and medium enterprise.
So households and businesses accumulate wealth. They invest in their own business or save through these financial intermediary mechanisms. Through the lens of the model and in the data, we see this hump shaped income Gini, and it has already been anticipated. This should not be regarded as a surprise. The Kuznets curve is well documented in many countries. Here it's working through the opening of branches for unbanked populations, which increases financial access, a relatively small number of talented entrepreneurs gain access to credit and more income, increasing inequality. There is a demand for labor by those firms, but it's gradual and wages tend to rise later, not right away. There is eventually a shortage of workers in a significant wage take off. But as I said, it takes time to those firms to expand. Eventually, the Gini index of inequality falls. There are some unanticipated results from this first experiment or match with historical data. Namely, economy-wide wage goes down in the initial years. So policymakers are looking at the data and see a drop in a wage. They may be quite alarmed.
Now, there's an easy explanation in the model. Farmers are becoming workers to supply the labor for these expanding firms, but initially, there's lots of labor and not so much demand. So the wage stays at or close to the wage of agriculture while workers make this transition. Observationally, it’s just a composition effect. You're adding more low income, low wage workers into the labor force. There's a lot of spatial heterogeneity. The local wage take offs occur at different points in time and space. So there's going to be substantial cross market income inequality among the workers. Large welfare gains do happen for workers, but the shared prosperity takes some time, and the orders of magnitude and so on, are different. So it's a more nuanced picture.
Now we introduce digital banking. We asked this question, what if counterfactual, instead of the expansion of branches, way back in 1986 they had introduced digital technology, which lowers the spatial access costs to zero. But we're going to keep the collateral constraint. We can decompose that experiment into two parts or two channels. One household saving. There's paper currency in the model so replacing part or all of it with interest bearing accounts for the saving side, while letting the branch expansion reduce the spatial access costs as before. So this eliminates currency from the household wealth portfolio, which had been held for consumption smoothing, that increases welfare.
The second part, comparing the lowering of credit access costs immediately to zero in the initial year while letting the branch expansion reduce the deposit costs slowly as in the baseline with the expansion of the banks. This removes the use of currency by entrepreneurs for self financing of the capital that they wish to buy.
Quantitatively at the calibrated values, the credit effect is dominating the savings effect. But I hasten to add that there's complementarity in the sense that the funds for credit are coming from the savings side through the intermediation. One cannot literally and would not want to do one side with without the other.
Now, the unanticipated results of the digital banking experiment, it does remove much of the regional heterogeneity. It lifts GDP and TFP, but the rise in those variables is slow. So there's still, despite lowering spatial access costs to zero, there's still underlying dynamics in the model. Surprisingly, to us, the income Gini is actually higher, almost right from the get go and sharply increasing. It’s “worse” relative to the expansion of historical expansion of branches. This is due to both the wage effect and the profit effect. The wage effect is similar to what I said before. Namely, there's an increase in demand for labor by the firms, but they're not getting paid very much. It takes about 5 years for the wage to actually be higher than it was in the baseline historical experiment with the branches. Entrepreneurs eventually accumulate wealth and their demand eventually increases the wages.
On the profit side, the profits of entrepreneurs increased dramatically with digital banking. This drives up, in fact, given their demand for credit, the interest rate significantly, so that only the productive entrepreneurs survive. So we have kind of a very large selection effect. The not so productive entrepreneurs quit their business, because the interest rates are high, and in turn, become workers or even go back to the farm. Over time of course entrepreneurs accumulate more and more wealth, savings is higher and the interest rates decline, which reverse this effect.
That model ignores something potentially quite important with which is risk and role of constraints and insurance and in turn related to intergenerational inequality. So in this paper, with Dejanir Silva, we look at risk taking over the life cycle. Again we “validate” the model. The model is one of risk-taking entrepreneurs who face limited, substantial, but still limited insurance against idiosyncratic shocks. The model successfully accounts for the idiosyncratic risk premium inequality and life cycle patterns of consumption and risk taking in the data. We can conduct two distinct counterfactual reforms. And here I want to make the point that it's important to distinguish the various aspects of digital technologies and smart contracts. One reform reduces risk constraints due to better pooling among strangers on a platform, which leads to a reduction in the idiosyncratic risk premium and leads to an investment boom. The initial generation of entrepreneurs benefit from this better insurance, but future generations are actually worse off as a consequence of the reform.
The second reform allows labor income from other members of the household to be committed to human capital as collateral on the distributed ledger, because you can book those future wage contracts and then lend against it. So this increase in borrowing impacts life cycle investment and inequality dynamics. It does so differently than relaxing the physical collateral constraints. Hopefully, the model is sort of making us more aware of these distinctions.
My final slide is to review in a way there is various obstacles and there's great variation across countries and all these things matter for policy. So in this paper with Dabla-Norris, Ji and Feliz-Unsal, we looked at multiple financing frictions: financial access, collateral constraints, bank markups due to verification costs. Each of those separately can be addressed by different aspects, of the, digital technology, lower physical access I talked about in the first model, data history or information is a partial substitute for monitoring costs, would lower the spread and finally, lower collateral needed, as I just discussed, in the risk life cycle context.
So our conclusion in this paper is the most binding constraint varies across countries. It's different in different countries. The type of constraint that's most binding is not obvious from the data. One needs to look at the data through the lens of the model. It's possible to get it wrong, relaxing the collateral constraint may indeed raise GDP in one country, but it may not be effective in other countries. As financial inclusion turns out to be more constrained by spreads or higher access costs that were not evident, at least not initially. This kind of research is very relevant for policy. It's not “just an academic exercise”. We develop this model in conjunction with the IMF and it was also used by the IADB at the level of country desk and actually used an official IMF guidance. And policy folks seemed to know that they couldn't have the Apple pie and there are tradeoffs. So they were actually quite attracted by the model allowing them to think through these kinds of tradeoffs.
For more information, please visit Luohan Academy's youtube channel: Luohan Academy