Sofonias Alemu Korsaye

We propose a model-free methodology to estimate international stochastic discount factors (SDFs) that jointly price cross-sections of international stocks, bonds, and currencies in markets with frictions. We theoretically establish a SDF decomposition into one global factor and a currency basket. We show that our global factor prices a large cross-section of international asset returns, not just in- but also out-of-sample, across different currency denominations. Moreover, the pricing ability of the global factor is largely independent of the market structure or the size and type of market friction.

We provide a no-arbitrage framework for stochastic discount factors (SDFs) that satisfy convex pricing constraints in markets characterized by a wide range of trading frictions. We demonstrate a duality relationship connecting minimum dispersion SDFs to portfolio optimization problems with penalty functions directly capturing the underlying frictions. Empirically, we examine how mispricing impacts the SDF’s effectiveness in explaining both cross-sectional and time series variation in asset returns. We find that a minimum-variance SDF, constructed by combining the capital asset pricing model SDF with a portfolio that constrains the mispricing of nonmarket risks, achieves a favorable tradeoff between time series and cross-sectional fit.

I develop a theoretical framework to identify investors’ subjective beliefs that are jointly consistent with survey expectations and asset prices in markets where investors face trading frictions. To quantify the deviation of investors’ beliefs from Rational Expectations (RE), I provide a metric that is interpretable as an upper bound on the difference between the maximum Sharpe Ratios attainable under investors’ beliefs and RE. Empirically, I show that (i) an economically significant share of the deviation of investors’ beliefs from RE, assessed assuming frictionless markets, can be attributed to small amounts of trading costs, and (ii) the size of the deviation from RE, as well as the share of the deviation attributable to trading costs, varies across investor and asset characteristics. More specifically, the expectations of sophisticated investors are closer to RE and more in line with asset prices.

Forecasting individual stock returns is characterized by a particularly low signal-to-noise ratio. We argue that the conventional zero-return forecast is an overly permissive benchmark because improvements over it can arise from predicting aggregate market movements rather than cross-sectional return differences. We propose the panel historical mean as a more informative benchmark, with an interpretation as the historical average return of the market portfolio. This benchmark is also nested in conventional machine-learning specifications when firm characteristics contain no predictive information. We study the challenge of improving upon this benchmark using a finite-sample analysis of Ridge regression under weak predictive signals. Empirically, consistent improvements relative to the panel historical mean arise only from flexible nonlinear methods and are concentrated among smaller, value-oriented, and less liquid stocks.

Instrumented Principal Component Analysis (IPCA) extends latent factor estimation by exploiting observable characteristics in addition to the underlying panel data. The method models the relationship between time-varying instruments and factor loadings, rather than imposing a separate fixed loading for every cross-sectional unit. This approach improves estimation when factor loadings evolve over time and the cross section is large, while helping connect latent factors to observable economic information. IPCA is computationally efficient and accommodates unbalanced panels. We establish consistency and asymptotic normality under general assumptions and illustrate the method's performance through simulation exercises and empirical applications in equity asset pricing and international macroeconomics.