[Research]

Energy & Environment

Periods of widespread adoption of new technology (“technological revolutions”) are important to understanding broader economic growth but diffusion does not receive as much attention in the literature as invention. This paper contributes to this question by studying a recent technology revolution, fracking in the American oil & gas industry. I present empirical evidence that shared knowledge networks which arise when production information cannot be perfectly excluded from other firms results in productivity improvements when aggregate investment activity increases. Firms are aware of this valuable intangible capital and respond to other firms’ investment decisions by shifting their own investments towards the new technology in areas where network effects are strong. I develop a novel empirical framework which uses a spatial panel model to first estimate network strengths. Then I stress test that network in a 2SLS specification with plausibly exogenous instruments for industry-wide investment levels. I use a rich dataset to measure a new variable, firm-level adoption rates of new technology. Cross-sectional differences in the usefulness of shared knowledge networks shows that network effects do not have as meaningful of an impact on how much firms invest as they do on the adoption process or how they allocate investments between old and new technology.

The plot shows the coefficient from regressions of productibvity on instrumented investment activity by other nearby firms. As you move up the quartiles of estimated knowledge networks along the x-axis, the relationship between productivity and ex-firm investment increases significantly.

Aggregate Technological Change: The joint effect of firm heterogeneity & knowledge spillovers

In the presence of knowledge spillovers, the distribution of firm types becomes an important input in the process of aggregate technology improvement. Firms are heterogeneous in where they choose to invest as well as how much they are able to invest. Using recentered influence function regressions, I provide empirical evidence for these differences in the American oil & gas industry. I show the effect of changing the distribution of firms on each part of the technology distribution and compare that effect across regions with different knowledge sharing propensity. The higher quantiles of the technology distribution is shown to be positively affected by distributions with higher average firm skills while the lower quantiles are negatively impacted. On the other hand, when the average firm size is higher, the median part of the technology distribution is positively impacted while both the quantiles on either end are either negative or neutrally impacted by the distribution of firm size.

Innovation & Growth

Shared Intangibles & Technology Adoption: A heterogeneous firm model

I build a heterogeneous firm model of technology adoption. Firms have the choice to produce a single final good using two different types of capital. Firms jointly solve an investment problem as well as stopping time problem which determines their technology specialization rate. The key spillover mechanism is captured in the function governing the productivity of new type capital which includes both the firm's own experience and the aggregate industry-wide experience as an input. I study the comparative statistics for economies with different knowledge sharing and learning propensities. I show that the spillover mechanism alone can generate the gradual adoption patterns often observed empirically. At the same time, learnng by doing within firms does not result in the same transition patterns.


(coming soon, preliminary draft available upon request)
The plot shows transition dynamics for an economy that starts with very few firms using the new type technology. The different lines show comparative statics of economies with dfferent shared knowledge and own learning propensities.

Household & Real Estate

Do Intermediary Constraints Matter? Evidence from Household Finance

(available upon request)

Using servicers in the mortgage backed security market, I show that constraints placed on intermediaries through a specific channel, mortgage advances, has a statistically significant impact on loan outcomes even after controlling for borrower factors like credit score. Using a loan level dataset, advance behavior is shown to affect foreclosure timelines, a previously unexplored area. The probability of a mortgage resolving through borrower payoff rather than foreclosure, as well as the loss amount on liquidated mortgages is also shown to be affected by differences in the servicer's advance propensity.

The Image shows the the ratio of advances conditional on rolling over into 90 dpd and distress between bank and non-bank servicers. Starting in 2010, the probability that a bank chooses to advance conditional on a homeowner entering distress diverges between the bank types.

Other Work

Robust Regimes: the direct impact of model uncertainty

Hansen-Sargent models of robustness consider agents who are concerned with model misspecification. Detection error probabilities are sometimes used to train the amount of uncertainty with which agents should be worried. I propose a method to empirically estimate the impact of model misspecification on decisions. For a set of models with distorted transition matrices relative to the estimated (approximating) model, I show that the set of models which are statistically difficult to distinguish is not monotonic in the size of the distortions applied. For any level of detection error tolerated, forecasts which incorporate all models in that set outperform the approximating model regardless of the distortion size. Finally, the larger the forecast differences between the approximating and distorted models, the larger the impact on the probability that firms execute on long term real options. By contrast, short term decisions do not show as strong of an effect when the set of indistinguishable models imply smaller forecast differences. The method directly incorporates not just potential model misspecification but the forecast implications between models.

vera.chau@unige.ch