N°26-20: Large and Deep Factor Models

We show that a deep neural network (DNN) trained for stochastic discount factor (SDF) estimation admits a sharp additive decomposition that separates characteristics discovery from pricing rule estimation. The pricing-relevant component of this decomposition is governed by a new object, the Portfolio Tangent Kernel (PTK), which summarizes the features learned by the network and yields a linear factor representation for the SDF. In population, the PTKimplied SDF converges to a ridge-regularized version of the true SDF, where the effective degree of regularization is determined by the spectral complexity of the PTK. Using U.S. equity data, we show that the PTK representation delivers economically and statistically significant gains in SDF performance, but its spectral complexity has increased sharply over time-by roughly a factor of six since the early 2000s-coinciding with a deterioration in finite-sample pricing performance.