From Fundamental to Quantamental Investing

From Fundamental to Quantamental Investing

Not a technical post. Just some high-level thoughts on how to pivot a hedge fund (or other financial institutions) from fundamental investing to a more quantitative approach.

  • Don’t invest in crazy scientists right away. Bring first strong data engineers: Many of them can also deliver very decent Machine Learning pipelines and Data Science studies. You really want someone with infrastructure/database/software engineering skills to start the journey towards applications of Data Science & Machine Learning for your business.

  • You should keep focus on your core expertise (e.g. fundamental investing), and use Machine Learning to automate tasks (e.g. data entry and collection, alerts, and other mundane repetitive tasks an entry-level analyst would do). Doing so, you will free more time for interesting value-adding (fundamental) research. Then, use Data Science / Machine Learning to progressively add incremental value to existing strategies (e.g. recommendation of opportunities based on similar past trades, estimation of (conditional) probabilities for scenarios foreseen by the portfolio manager with probabilistic graphical models).

How I would structure such an effort:

  1. Take a portfolio manager (PM) who masters his/her game, has already a good running strategy, and is willing to collaborate to make his/her pod more efficient, and eventually increase the P&L of his/her business line.

  2. Associate to the PM a well-rounded data engineer.

  3. Gather the relevant data, and build a clean database for the PM (here, some Machine Learning may already be involved).

  4. Build analytics (e.g. inside a modern web-app) for the PM so that he can drill down in the data, looking for clues, testing investment hypotheses, etc. You can potentially pepper a few predictive analytics here and there (e.g. regression results, features importance) to give the PM a clue how a Machine Learning model would leverage the available data: Would it match the intuition of the PM on which data points and features are important?

  5. At stage 4, you already have something which should be useful and value-adding for the PM (otherwise he/she poorly specified the needs). If you can or want to invest more, you may want to systematize/automate the strategy: Typically, systematizing the strategy will lower the hit rate compared to the good PM’s one, but it will allow for a higher coverage (since there is only a marginal cost for considering more opportunities) compared to what the good PM can cover (with a couple of analysts). If successful, it can be a strategy running alongside the PM’s one, and eventually executed (if needed) by a junior PM/trader.

Then, rinse and repeat. If you manage to successfully unroll these steps with a couple of portfolio managers, and they are happy with the results, you may consider to bring in a central team of data engineers and scientists. Doing this approach a couple of times on particular cases will help you to avoid the common error of setting-up a central team which develop “generic” frameworks which should be useful to everyone but are really helping no one.

At this stage, the business should have a couple of portfolio managers, analysts, and engineers who know how to work together efficiently.

If you decide to go forward with a central Data Science & Machine Learning team:

  1. Once again, start first with the data engineers who can set up infrastructure, basic Machine Learning pipelines, and dashboards, which will be shared across the business this time.

  2. Hire top scientists (which are specialists) only when basic data and analytics pipelines are already running, adding value, and you want to improve on an already successful framework with more sophistication (which may not necessarily improve on the baseline).

These recommendations do not hold if you are in the top tier systematic quant trading business, obviously.