Research

Job Market Paper

Is the Dollar Reconnect Forecastable? Information Timing and Exchange-Rate Predictability

This paper asks whether the post-2007 reconnect between the U.S. dollar and global financial variables translates into real-time exchange-rate predictability.

Using 13 USD exchange rates with out-of-sample evaluation through December 2025, I compare two information sets for the same one-month exchange-rate return: a feasible forecast-origin model using the S&P 500 return available when the forecast is formed, and a realized-regressor benchmark using the S&P 500 return realized over the forecast window. The realized-regressor benchmark is not a feasible forecast; it measures out-of-sample fit using information observed only after the fact.

The difference is large. With realized equity information admitted, out-of-sample fit is strong and rises sharply after 2007, consistent with the documented dollar reconnect. At the forecast origin, performance remains weak and negative on average. The results therefore show that much of the apparent forecasting gain associated with the reconnect comes from information realized during the forecast window rather than information available to the forecaster in real time.

The paper also examines alternative estimators and predictor sets, including OLS, LASSO, random forest, and XGBoost, and finds that weak forecast-origin performance is robust across specifications.

Keywords: Exchange rates, forecasting, information timing, real-time information, out-of-sample predictability

Paper PDF — coming soon


Dissertation

When Long-Run Evidence Fails: Adjudication, Failure Modes, and Detectability

This chapter examines whether exchange rates exhibit stable long-run relationships with macroeconomic fundamentals once integration properties, structural breaks, deterministic specifications, and model admissibility are treated as part of the empirical evidence.

Across 14 USD exchange rates, no currency obtains primary support for a common four-variable long-run specification. The absence of support arises through several distinct empirical channels, including direct lack of cointegration evidence, integration ambiguity, deterministic-specification sensitivity, and dynamic misspecification.

The chapter also studies whether slow economic adjustment may make economically meaningful relationships difficult to detect in finite samples.


Unit of Analysis and Pooled Identification

This chapter asks whether pooling information across currencies can recover structural or predictive content that is difficult to identify in individual exchange-rate models.

The analysis changes the unit of estimation while holding the underlying data construction fixed, providing a bridge between the long-run and forecasting evidence in the preceding chapters.

Work in progress.


Publications

Why Has the United States Continued to Receive Net Investment Income as a Debtor Country?

with Juann Hung

The World Economy, 42(3), 2019.

Journal Link


Other Research

Exchange Rate Forecastability with Monetary Policy Rule Models and Machine Learning

with Merih Uctum

Compares purchasing power parity, uncovered interest parity, and Taylor-rule fundamentals with machine-learning and forecast-combination approaches.

Sources of Advanced Economies’ Outward Foreign Direct Investment Return Advantage

with Juann Hung

Examines the sources of cross-country differences in outward foreign direct investment returns using time-series and difference-in-differences evidence.