Perceived vs. Revealed Risk Tolerance For Efficient Asset Allocation

Abstract

Risk tolerance is a key parameter in asset allocation and subsequent portfolio management. The risk tolerance is usually estimated using the investor's ability (age,wealth etc.) and willingness to take risk. However, investors are subject to many behavioral biases and their risk perception varies significantly with the market. This paper uses the actual data of the investors and establishes the fact that investors are poor judges of their own risk aversion and their risk tolerance changes with the recent market movement. The paper then discusses the methodology to compute or correct the risk tolerance of the investors using the behavior of similar investors in different market condition. Machine learning techniques are used to identify the clusters of investors with significant change in risk tolerance and underlying factors driving the change. The asset allocation with this method of computing risk tolerance is more likely to maximize the utility of an investor. Also, having a portfolio consistent with their true risk profile will let the investors stick to their long run strategy without overreacting.

Keywords: Risk Tolerance, Relative Risk Aversion, k-means, Principal Component Analysis, Asset Allocation, Random Forest, Gradient Boosted Trees