Machine learning methods for big data trade off bias for precision in prediction. To understand the implications for financial markets, I formulate a trading model with a prediction technology where investors optimally choose a biased estimator. The model identifies a novel cost of complexity that arises endogenously. This effect makes it optimal to ignore costless signals and introduces in- and out-of-sample return predictability that is not driven by priced risk or behavioral biases. Empirically, the model can explain patterns of vanishing predictability of the equity risk premium. The model calibration is consistent with a technological shift following the rise of private computers and the invention of the internet. When allowing for heterogeneity in information between agents, complexity drives a wedge between the private and social value of data and lowers price informativeness. Estimation errors generate short-term price reversals similar to liquidity demand.

This paper develops an asset market-based test for preference for the timing of resolution of uncertainty. Our main theorem provides a characterization of preference for early resolution of uncertainty in terms of the risk premium of assets realized during the period when the informativeness of macroeconomic announcements is resolved. Empirically, we find support for preference for early resolution of uncertainty based on evidence on the dynamics of the implied volatility of S&P 500 index options before FOMC announcements.

I examine the implications of time inconsistency, modeled by hyperbolic discounting, for the excessive trading puzzle and asset prices. I show that unlike the case of long-term contracting with naive time-inconsistent agents where the welfare inefficiency of naivete disappears, dynamic trading allows time-consistent agents to exploit naive agents even over long horizons. In addition, partial awareness of the naivete bias induces leading trading motives such as gambling behavior and perceived information advantage, which can serve as a microfoundation for the puzzling excessive trading volume observed empirically. I show that the asymmetric information about the extent of partial naivete creates uncertainty about the optimal trading contract that the time-consistent agent can offer and endogenously generates extra risks in her consumption dynamics. As a result, the presence of naive time-inconsistent investors increases the risk-free rate, volatility, and risk premium in the economy.