Unlocking Hidden Nonlinearities: A New Approach to Stock Return Predictability

The Puzzle of Predictability

Have you ever wondered why some investors seem to consistently beat the market while others struggle? One reason might be their ability to predict excess stock returns based on past data. While it's widely believed that excess stock returns exhibit a certain degree of predictability over time, traditional predictive regressions often detect this predictability as statistically small. However, the direction-of-change and volatility of returns show a substantially larger dependence over time.

In this blog post, we'll analyze an innovative approach to predicting excess stock returns by modeling individual multiplicative components and combining their information to recover the conditional expectation of the original variable of interest. This method is detailed in the working paper "Modeling Financial Return Dynamics via Decomposition" by Stanislav Anatolyev and Nikolay Gospodinov.

The Intriguing Decomposition

The return at period t (rt) can be factored into two multiplicative components: |rt|, the absolute value of rt, and sign(rt), the sign of rt. This decomposition, called an "intriguing decomposition" in Christoffersen and Diebold (2006), allows for a more nuanced understanding of return dynamics.

Anatolyev and Gospodinov's approach involves jointly modeling the absolute values and signs of returns to better capture hidden nonlinearities in excess return dynamics that cannot be captured in standard predictive regression setups. By using a multiplicative error model for absolute values, a dynamic binary choice model for signs, and a copula for their interaction, this method can pin down the conditional expectation Et−1 (rt).

Volatility Persistence and Sign Predictability

Volatility persistence and predictability have been extensively studied in the literature. For example, Andersen et al. (2006) documented volatility persistence and predictability using absolute values of returns. Meanwhile, researchers like Christoffersen and Diebold (2006), Hong and Chung (2003), and Linton and Whang (2