Lecture 13: Portfolio Management

Quick Overview

Portfolio management fundamentally involves determining the correct sizing of investments based on clarified objectives, loss tolerance, and acknowledging that volatility is an imperfect measure of risk, leading the speaker to propose using an expected gain-loss ratio as a superior sizing metric over traditional methods like Sharpe ratio.

Key Points: Portfolio construction is primarily a sizing problem, determining the relative amounts of chosen investments to meet objectives while minimizing risk, which the speaker emphasizes is often misinterpreted using volatility. The classic Modern Portfolio Theory (MPT), pioneered by Harry Markowitz, relies heavily on capital market assumptions (return, volatility, correlation) which are hard to predict accurately in practice and lead to unstable optimization solutions. The speaker introduces a new sizing metric, the Gain-Loss Ratio (G/L ratio), derived from Kelly's criteria, which translates directly into sizing percentage and is bounded from -1 to +1, aiming to maximize expected gain (G) while minimizing expected loss (L). Diversification provides a 'free lunch,' but only if the portfolio is rebalanced, as demonstrated by an example where perfectly negatively correlated assets yielded 0% return over two years without rebalancing, but 25% compounded annually with rebalancing. Endowment funds typically target an 8% nominal return, needing 5% for spending plus 3% for inflation, relying heavily on external managers for both public and private investments. Crowding behavior, where agents react to observations and amplify each other's actions (like synchronized walking on the Millennium Bridge), can drive markets to unstable extremes, leading to bubbles and crashes, which is a core concept in behavioral finance. Power law distributions, characterized by scale-free properties (e.g., 20% of people having 80% of wealth), originate from these crowd interaction feedback loops where agents with more power gain more power, unlike natural phenomena like human height distribution.