Serhiy Kozak

Associate Professor of Finance at the University of Maryland's Robert H. Smith School of Business

UMD Scholar LinkedIn GitHub

Download CV

Latest Publications

Recent contributions to top-tier academic journals advancing our understanding of financial markets and asset pricing

Equity Term Structures without Dividend Strips Data (2024)

with Stefano Giglio and Bryan Kelly

Journal of Finance 79(6), 4143-4196

PDF JF SSRN Appendix SlidesCode

citations

Equity Term Structures Without Dividend Strips Data: A Novel Approach to Asset Pricing

This paper develops an affine model to estimate the term structures of equity discount rates and dividend yields for a broad cross-section of portfolios, using only equity returns and dividend data—without relying on dividend strip data. The model accurately replicates observed dividend strips and generates extended term structures over longer time spans and across diverse equity portfolios. These results provide a new set of empirical moments to evaluate asset pricing models.

Key Contributions

Innovative Methodology: We propose an affine model driven by a state vector including four factor returns and dividend-price ratios. This approach captures the dynamics of discount rates and dividend yields without relying on dividend strip markets.
Cross-Sectional Term Structures: The model produces term structures for 102 portfolios, such as size, value, and momentum portfolios, highlighting heterogeneous patterns across equity strategies.
Validation Against Observed Strips: The model's implied term structures closely match traded dividend strips in terms of levels, slopes, and time-series behavior, providing strong out-of-sample validation.

Empirical Findings

For the S&P 500, the model replicates key features of the term structure observed in traded dividend strips, such as mild upward slopes in normal times and inversions during recessions.
The term structures of portfolios like value, growth, small, and large stocks exhibit distinct slopes and time-series behaviors.
The term structure inverts during recessions across many portfolios, reflecting increased risk premia for short-horizon claims during economic downturns.

Why It Matters

The rich cross-section of term structures provides new empirical moments to calibrate and test structural asset pricing models, including implications for long-run risk and habit formation models. The model enables valuation of projects with specific maturities, such as climate change investments or private equity, and facilitates the construction of hedging portfolios aligned with specific risks.

This paper advances the study of equity term structures by offering a scalable, data-driven approach that expands the horizons of empirical research and practical applications in finance.

Dynamics of Bond and Stock Returns (2022)

Journal of Monetary Economics 126, 188-209

PDF JME SSRN Appendix

citations

Dynamics of Bond and Stock Returns: Understanding Time-Varying Correlations

The 2008 financial crisis highlighted the role of US Treasury bonds as a hedge against stock market risk, creating episodes of negative bond-stock correlation. Conversely, the 1990s were characterized by positive correlations. Existing models have struggled to explain these patterns. This paper addresses the limitations of such models by introducing a framework that combines time-varying risk aversion with dual-production technologies, enabling endogenous fluctuations in bond-stock correlations and term premia.

Key Contributions

Dynamic Risk Mechanisms: The model incorporates two key mechanisms - technology diversification with varying cash-flow risks, and flight-to-safety driven by time-varying risk aversion during crises.
Realistic Bond and Stock Pricing: The model endogenously generates term premia and bond-stock correlations that change sign, reflecting observed market patterns.
Empirical Validation: Model predictions align with key empirical phenomena, such as the relationship between bond-stock correlations, risk aversion proxies (e.g., VIX, credit spreads), and yield curve dynamics.

Empirical Findings

Historical data shows that bond-stock correlations co-move with measures of risk aversion, such as market volatility and credit spreads.
Bond-stock correlations effectively predict variations in bond and equity risk premia, with high correlations corresponding to increased bond risk premia.
Changes in bond-stock correlations significantly influence the slope and level of the yield curve, consistent with model predictions.

Why It Matters

The findings highlight the importance of accounting for dynamic correlations in portfolio construction, especially during periods of heightened market stress. The model provides a framework for understanding how macroeconomic shocks affect financial markets through risk aversion and investment reallocation mechanisms.

This paper presents a robust framework for analyzing the joint dynamics of bonds and stocks, offering new insights into the interplay of risk aversion, investment reallocation, and market pricing.

Factor Timing (2020)

with Valentin Haddad and Shrihari Santosh

Review of Financial Studies 33(5), 1980-2018

2018 Q-Group Jack Treynor Prize

PDF RFS SSRN Appendix

citations

Factor Timing: Enhancing Portfolio Performance Through Dynamic Strategies

This paper investigates the potential of factor timing strategies to improve portfolio performance. We show that optimal factor timing portfolios align with the stochastic discount factor (SDF), leading to significant benefits compared to static factor investing. By leveraging the predictability of equity factor returns, particularly the principal components of anomalies, our study reveals robust improvements in Sharpe ratios and offers new insights into the time-varying dynamics of risk premia.

Key Contributions

Dynamic SDF Estimation: We propose a method that incorporates the predictability of factors, focusing on the largest principal components (PCs) to reduce dimensionality and stabilize forecasts. This approach ensures an economically plausible and robust SDF.
Predictability of Factor Returns: The study demonstrates that several PCs of equity anomaly portfolios, particularly the first and fourth components, exhibit strong out-of-sample predictability using their own valuation ratios (e.g., book-to-market).
Portfolio Performance: Factor timing strategies significantly enhance performance, achieving higher Sharpe ratios than static factor investing. A pure factor timing portfolio achieves a Sharpe ratio of 0.77 out of sample, rivaling traditional market timing.

Empirical Findings

Using 50 equity anomaly portfolios from 1974-2017, we find strong evidence of factor return predictability.
The first principal component explains 25.8% of the variance in anomaly returns, with predictability outperforming aggregate market returns.
Factor timing provides substantial economic benefits, with significant increases in expected utility and Sharpe ratios compared to static strategies.

Why It Matters

The findings challenge traditional SDF models, highlighting the importance of incorporating time-varying factor risk premia into asset pricing frameworks. Our factor timing strategies can be implemented in practice, offering investors a way to enhance returns without relying solely on static exposures.

This paper advances our understanding of dynamic portfolio strategies and presents a compelling case for incorporating factor timing into both academic and practical portfolio construction frameworks.

View All Publications

Latest Working Papers

Current research projects and papers under review, exploring cutting-edge topics in financial economics

When do Cross-Sectional Asset Pricing Factors Span the Stochastic Discount Factor? (2024)

with Stefan Nagel

PDF SSRNSlides

citations

When Do Cross-Sectional Asset Pricing Factors Span the SDF? A Framework for Factor Construction

This paper explores the conditions under which cross-sectional asset pricing factors constructed from firm characteristics span the Stochastic Discount Factor (SDF). We show that commonly used heuristic factor construction methods—such as sorting, weighting by characteristics, or OLS regression slopes—only span the SDF under specific structural conditions on the covariance matrix of asset returns. We propose methods to relax these conditions, including hedging unpriced risks and applying dimensionality reduction techniques.

Key Contributions

Conditions for Spanning the SDF: Factors span the SDF if the covariance matrix of asset returns satisfies a specific structure where characteristics explain systematic risk and orthogonalize unpriced risks. This structure is more likely to hold when using a large number of characteristics.
Hedging Unpriced Risks: We introduce a hedging procedure to remove unpriced risk exposures from heuristic factors, iteratively improving their alignment with the SDF.
Dimensionality Reduction: We identify conditions under which the dimensionality of characteristics can be reduced without losing critical pricing information. Techniques like IPCA and PPCA are shown to span the SDF effectively.

Empirical Findings

Hedged factors consistently achieve higher Sharpe ratios compared to unhedged factors, particularly in small-scale models with fewer characteristics.
Factors constructed using GLS or hedging outperform traditional heuristic factors in capturing the SDF, especially when the number of characteristics is small.
IPCA and PPCA, when applied under the right covariance matrix conditions, effectively reduce the number of factors needed while maintaining pricing efficiency.

Why It Matters

Incorporating hedging or GLS-based approaches into factor construction can significantly enhance the efficiency of asset pricing models. Using a comprehensive set of characteristics improves the likelihood that factors span the SDF, even if some characteristics do not directly relate to expected returns.

This paper delivers a rigorous theoretical and empirical analysis of factor construction methods, offering actionable insights for refining cross-sectional asset pricing models.

Kernel Trick for the Cross-Section (2023)

PDF SSRN

citations

Kernel Trick for the Cross-Section: Incorporating Non-Linear Relationships in Asset Pricing

Traditional characteristics-based asset pricing models, such as Fama-French, assume that expected returns are linear functions of pre-defined characteristics. This simplification ignores critical non-linearities and interactions that theory and accounting identities suggest should matter. This paper addresses these limitations by using the "kernel trick" to extend characteristics into a high-dimensional, potentially infinite-dimensional space. This method allows for the inclusion of all interactions and non-linear functions without incurring prohibitive computational costs.

Key Contributions

Introduces a kernelized PCA (CK-PCA) framework to capture non-linear relationships among characteristics, utilizing economically motivated regularization rooted in Bayesian priors
The kernel trick enables transformation of characteristics into a high-dimensional space without explicitly calculating features, mitigating computational burden
Supports inclusion of higher-order interactions and non-linear terms, extending factor model flexibility
Demonstrates that allowing for non-linearities doubles the out-of-sample Sharpe ratios of MVE portfolios

Empirical Findings

The radial kernel improves Sharpe ratios from 1.65 (linear model) to over 3.0 for anomaly-based factors
Results hold in out-of-sample tests with significantly higher predictive power for firm-level and aggregate market returns
Kernel-based methods reveal critical features like higher-order momentum terms that traditional models miss
Conditional expected returns explain up to 0.8% of firm-level return variations, a fourfold improvement over benchmarks

Why It Matters

The paper makes a compelling case for integrating machine learning techniques into financial economics. By bridging theoretical insights with computational innovations, it provides a scalable method to handle high-dimensional data and suggests that non-linear factors are essential for capturing the true cross-section of expected returns.

The study demonstrates that incorporating non-linearities and interactions through kernel methods significantly enhances asset pricing models, leading to more efficient portfolios and better return predictions while providing practical tools for portfolio management.

Banking Deregulation and Stock Market Participation (2022)

with Denis Sosyura

PDF SSRN

citations

Access to Credit and Stock Market Participation: How Banking Deregulation Shapes Household Investment

The transformation of the U.S. banking sector via interstate deregulation enabled national banks to expand their reach, profoundly affecting household financial behavior. This paper explores the causal relationship between access to credit and two critical financial decisions: stock market participation and portfolio allocation. The research utilizes staggered state-level deregulations to isolate the effects of increased credit supply on these household choices.

Key Contributions

Uses staggered implementation of interstate banking laws as an exogenous instrument to measure credit supply impact
Leverages proprietary datasets on credit histories and brokerage accounts for robust micro- and macro-level insights
Demonstrates a causal link between credit expansion and increased stock market participation
Identifies demographic and geographic heterogeneities in effects across household types

Empirical Findings

Deregulation led to a 34% increase in credit card issuance within two years, followed by a 13% rise in stock market participation
Younger and credit-constrained households disproportionately benefited, increasing risk tolerance and shifting from cash to equities
Geographic analysis revealed localized effects near new bank branches, emphasizing physical credit supply's role
Results robust to controls for local economic factors, migration, and pre-existing trends

Why It Matters

The study provides compelling evidence that access to credit plays a crucial role in household financial decisions, with implications for both policy and financial institutions. The findings challenge traditional models by emphasizing credit constraints in portfolio choices and explain the rise in market participation from 19% in the 1980s to over 30% in the 1990s.

This paper bridges gaps between household finance, banking, and capital markets, demonstrating how banking deregulation and increased credit access significantly influenced stock market participation, particularly benefiting younger and credit-constrained households.

View All Working Papers

Research Fields

Core areas of research focus, combining theoretical frameworks with empirical methodologies

Empirical Asset Pricing

Developing novel methodologies to understand patterns in asset returns. I create innovative statistical approaches to analyze large panels of returns and combine machine learning with traditional asset pricing theory to improve our understanding of risk premia.

Theoretical Asset Pricing

Investigating the fundamental principles that govern asset prices, with focus on systematic risk factors and stochastic discount factors. I develop frameworks that bridge theoretical asset pricing with empirical applications.

Machine Learning in Finance

Adapting and extending modern ML techniques to address unique challenges in finance. I develop approaches that combine the flexibility of machine learning with economic intuition to handle the low signal-to-noise ratio, limited time-series, and structural breaks of financial data.

Research Areas of Expertise

Exploring cutting-edge topics in financial economics through interdisciplinary approaches and innovative methodologies

New Methods for Risk-Return Trade-off

Developing innovative approaches to understand how asset prices incorporate cross-sectional information and market participants' expectations.

Equity Term Structures without Dividend Strips Data (2024)

Shrinking the Cross-Section (2020)

Interpreting Factor Models (2018)

When do Cross-Sectional Asset Pricing Factors Span the Stochastic Discount Factor? (2024)

Machine LearningFactor ModelsAsset Pricing

Machine Learning and AI in Finance

Pioneering the application of advanced machine learning techniques to solve complex financial problems and portfolio construction.

Factor Timing (2020)

Shrinking the Cross-Section (2020)

Interpreting Factor Models (2018)

Machine LearningAIHigh-Dimensional Methods

Macroeconomic Dynamics of Asset Prices

Investigating how asset prices and risk premia evolve over time and across asset classes, combining structural and empirical approaches.

Dynamics of Bond and Stock Returns (2022)

Factor Timing (2020)

Why Do Discount Rates Vary? (2020)

Macro-FinanceRisk PremiaMarket Dynamics

Asset Pricing Across Markets

Exploring interconnections between different asset classes and global markets, focusing on joint pricing of treasuries and equities.

Equity Term Structures without Dividend Strips Data (2024)

Dynamics of Bond and Stock Returns (2022)

When do Cross-Sectional Asset Pricing Factors Span the Stochastic Discount Factor? (2024)

Global MarketsCross-AssetEconomic Theory

Top 5 Most Cited Papers

Most influential research papers based on citation metrics

Learn More About My Research

Research Impact

Measuring the influence and reach of research contributions through citation metrics and scholarly impact

0

Total CitationsAcross all publications

0

h-indexImpact measure

0

Citations per YearAverage over last 5 years

0

Total PapersAll research works

View Detailed Research Metrics

Quick Links

Navigate through key sections to explore research publications, resources, and teaching materials

Next Section

Learn more about my background