Simple (portfolio) heuristics that make us smart(er) 03/20/2025 Giuseppe (Gappy) Paleologo Balyasny Asset Management Plan & Message ● Goal : help bridge the gap between book-level Mean-Variance Optimization and application-level MVO… ● … by proposing modifications that give a Minimum Viable Product in closed form ● Messages : Simple is beautiful. Simple is not easy. ● Requirements : basics of factor models; basics of Mean-Variance Optimization (MVO). See book recommendations at the end. ● Disclaimers : the opinions expressed in this talk are not necessarily those of my cat. Or my employer. Also, not investment advice. A Practical Problem 1. 2. 3. 4. 5. 6. You have "alphas": expected returns of a large set of assets They change fast enough, and you are rich enough, that you have to worry about transaction costs You have a primitive factor model, e.g., a simple 1-factor model, or a more complex non-commercial one (e.g., https://github.com/0xfdf/toraniko) Optimization is either too complex or it gives unintuitive results You want to generate a tradeable portfolio that is essentially correct N.B.: I wrote a book and discarded a lot of material. This talk is discarded material from the red book What People Do ● Real weird stuff, like this: exp. return avg. daily volume or this: est. daily volatility but with ceiling and floors on the parameters because otherwise the formula gives strange recommendations A Different Procedure First, orthogonalize your alphas to factors. Very standard procedure (e.g., see my coming book on quant investing). For example, for a simple market model with betas Because of orthogonalization, in the absence of transaction costs, the MVO only sees the diagonal idiosyncratic covariance matrix. Second, solve the MVO problem. Now it's separable. => (separable because cov.mat. is diagonal) We need a market impact model. Almgren-Chriss: estimate k, calibrate it or get from a reputable source Reasonable Asymptotics Exercise: what is the asymptotic size for lambda -> Inf? Extension #1: Initial Trading Positions We can extend to non-zero initial positions, at zero-cost [pun!]. Intuition: 1. 2. change of alpha change of starting point Extension #2: Alpha Uncertainty This is much more important an non-trivial. Assume We estimate tau from the Information Coefficient. It is usually high! (~1/IC): => => Summing Up 1. Orthogonalize. Always 2. Regularize. Always 3. Account for transaction costs. Also Alw… 4. A lot of good science & engineering is knowing what terms to keep and what terms to drop Coordinates, books X (finance): @__paleologo LinkedIn: www.linkedin.com/in/gappy/ Web: linktr.ee/paleologo Advanced Portfolio Management The Elements of Quantitative Investing The title of the talk is a homage to a great psychologist (B.Gigerenzer). Here is his book: Simple Heuristics That Make Us Smart