Working Papers

The Evolution of the Network of Common Intermediation (10 Stock Subsample) - This figure illustrates the natural experiments we exploit. In 2018, the Brazilian stock exchange assigns 2 or 3 (out of 9) market makers to each of 54 (out of around 400) stocks. In 2019, the Brazilian stock exchange assigns 2 to 5 (out of 13) market makers to each of 89 stocks. The nodes in the graph are the ticker symbols of 10 randomly selected program stocks. Nodes are connected if they have a common market maker. Gray edges indicate that the two stocks are connected during both market maker programs. Red edges indicate that stocks were connected during the 2018 program, but got disconnected. Blue edges indicate new connections created by the 2019 program. One result of the paper is that assigning a common market maker to a stock pair increases the co-movement of returns, liquidity, and volume.

"How to Increase the Liquidity and Stability of Financial Markets?," with Pedro Tremacoldi-Rossi (This is my upcoming JMP, draft coming soon!)

The Ben Bernanke Prize in Financial and Monetary Economics

Using unique, full-depth, message-level trade and quote data from the Brazilian stock exchange that reveal market participants' identities, and exploiting two rich market maker programs as natural experiments, this paper studies the behavior of algorithmic market makers and its implications for asset prices and market quality. First, we find that in normal times exchanges can increase market quality by encouraging voluntary liquidity provision. In contrast, imposing tight liquidity provision requirements constrains market makers and can decrease market quality. In crisis, however, we find that voluntary liquidity providers withdraw and mandatory intermediaries become the liquidity providers of last resort. This suggests that the optimal approach may combine incentives with requirements that only bind in crisis. Second, we document that market makers' cross-asset hedging behavior causes excess co-movement of returns, liquidity, and volume. This shows a trade-off between liquidity and excess co-movement.

The anomaly network - two anomalies are connected if they form an interaction strategy that generates statistically significant (correcting for multiple testing) alpha beyond standard asset pricing factors and the factors corresponding to the underlying individual anomalies.

"Interacting Anomalies" (with Karsten Müller) (SSRN) (more results and data available on the project website)

An extensive literature studies interactions of stock market anomalies using double-sorted portfolios. But given hundreds of known candidate anomalies, examining selected interactions is subject to a data mining critique. In this paper, we conduct a comprehensive analysis of all possible double-sorted portfolios constructed from 102 underlying anomalies. We find hundreds of statistically significant anomaly interactions, even after accounting for multiple hypothesis testing. An out-of-sample trading strategy that invests in the top backward-looking double-sort strategy generates equal-weighted (value-weighted) monthly average returns of 4% (2.7%) at an annualized Sharpe ratio of 2 (1.38), on par with state-of-the-art anomaly-based machine learning strategies.

The staggered start of HFT across markets - This graph shows HFT start dates across the markets in our sample.

We study how the informativeness of stock prices changes with the presence of high-frequency trading (HFT). Our estimate is based on the staggered start of HFT participation in a panel of international exchanges. With HFT presence, market prices are a less reliable predictor of future cash flows and investment, even more so for longer horizons. Further, firm-level idiosyncratic volatility decreases, and the holdings and trades by institutional investors deviate less from the market-capitalization weighted portfolio as a benchmark. Our results document that the informativeness of prices decreases subsequent to the start of HFT. These findings are consistent with theoretical models of HFTs’ ability to anticipate informed order flow, resulting in decreased incentives to acquire fundamental information.

The price impact of mutual fund flow-induced fire sales - this heatmap shows the stock return caused by a 1% demand shock generated by extreme surprise flows that are in the top x or bottom y percentile.

I propose a new method to isolate a plausibly exogenous component of mutual fund flows in order to estimate the price impact of fire sales. The method addresses a potential reverse causality problem: instead of mutual fund outflows inducing fire sales, which drive down prices, poor stock returns reduce mutual fund returns, which in turn trigger outflows. The solution is to construct a new instrument from high-frequency surprise flows. Using surprise flows to reexamine important findings in the literature, I find equity markets are deeper and less distortive than suggested.

Illustration of the identification strategy and findings for the case of cash dividends - Demand pressure effects differ from news effects in terms of timing; news change asset prices at announcement, while demand pressure changes asset prices at payment, when financial institutions receive cash and reinvest. I examine returns of the firm itself, peer stocks, and connected stocks around announcement and payment dates at the daily frequency. Dividend firms experience abnormal returns after both the announcement and the payment, as is well-known. Next, I find no evidence for spillovers on industry peers. Finally, connected stocks show payment date price pressure effects but no announcement date news effects.

This paper uses the reinvestment of corporate payouts by financial institutions as a nonfundamental shock to asset prices to estimate the slope of the demand curve for stocks and the real effects of stock returns on corporate financing and investment. Exploiting the separation of announcement and payment at the daily frequency, I find dividends in particular generate payment date price pressure but no announcement date news spillover effects, suggesting that dividend-induced trading is plausibly exogenous to fundamentals. Using dividend-induced trading as a natural experiment for stock returns, I estimate an asset demand elasticity of 1.25 and document a releveraging market feedback effect on investment, where firms respond to an exogenous stock price increase by issuing debt and use the funds to invest.

Work in Progress

Excess Returns of Portfolios Sorted on Select Home Characteristics - This figure shows percent average excess price appreciation of quintile portfolios sorted on continuous, micro-level property characteristics. The characteristics are sorted with respect to the return of the high portfolio. The sample covers 1996 to 2020.

"Risk and Return in the Residential Real Estate Market," with Mathias Büchner and Bryan Kelly

Using proprietary data from Zillow containing the universe of US real estate transactions from 1996 to 2020, we examine the cross-section of property-level returns. First, even after accounting for multiple hypothesis testing, we find many property characteristics predict returns. These include signals mirroring well-known anomalies in the stock market, such as value, size, momentum, long-term return reversal, and profitability. In addition, there are real estate-specific predictors: home age, the return realized by the last owner, local income per capita, housing supply growth, property taxes, and variables related to idiosyncratic risks. Second, observable factor models fail to price real estate assets. However, instrumented principle component analysis (IPCA) overcomes this issue: a model with two latent risk factors explains the cross-section of housing returns. Finally, using machine learning techniques, we document that real estate returns are highly predictable. Out-of-sample, houses in the top decile of predicted returns outperform homes in the bottom decile by around 10% annually, even after holding location, time, and size constant to account for investors' constraints.

Map of Total Climate Risk Aggregated to the County-Level - This figure shows percent expected annual losses due to risks that increase with climate change and directly affect real estate: wild fires, hurricanes, strong wind, coastal flooding, and riverine flooding.

"Real Estate Investing, Climate Risk, and Inequality"

This paper examines who bears climate risk via real estate holdings. I measure climate risk exposure as the expected annual loss to buildings using census-tract level climate risk data from FEMA. Further, I combine proprietary data from Zillow containing the universe of US real estate transactions from 1996 to 2020 with HMDA data which allows me to observe demographic information of buyers and sellers. This allows me to answer three questions: Who is most exposed to climate risks? How do real estate investors respond to climate change news? And how do asset prices react to climate change news? I find the top of the income distribution is most exposed to climate risks. However, in response to climate change news, the wealthy disproportionately sell exposed homes. These sales are absorbed by firms. Lastly, prices of exposed homes decline after climate change news.

We Track Investors on the Bitcoin Blockchain Using Known Addresses and Clustering Algorithms.

"Cryptocurrencies, Return Chasing, and Return Predictability," with Stefan Nagel and Zachry Wang

We combine the full information from the BTC blockchain with addresses of real entities and address clustering algorithms to construct a partially deanonymized BTC transaction history. We find that unsophisticated investors chase past BTC returns and that BTC purchases of sophisticated investors predict BTC returns.

Distribution of Out-Of-Sample AUCs by Predictor Variable Category - This figure shows boxplots of out-of-sample AUCs for a large set of candidate predictor variables by predictor variable category. In particular, we highlight the most prominent predictor of financial crisis from the literature in red; credit.

"Predicting Financial Crises: a Comprehensive Assessment," with Lu Liu and Karsten Müller

We provide a comprehensive evaluation of which variables are reliable predictors of financial crises. We evaluate thousands of candidate predictors and a large set of predictive models and find that variables related to credit, external imbalances, and macroeconomic conditions have robust forecasting ability out-of-sample.