Principles
What we hold to
- Survive.
- Preserve capital.
- Manage risk.
- Diversify.
- Rely primarily on simple strategies, such as trend following and mean reversion.
- Edges should be simple, mechanical, and repeatable.
- Try the “no brain” approach: let machine learning discover many uncorrelated alphas, and stitch them together.
- Monitor many strategies, around 200, but rotate only 20 or so into live trading each month, based on what’s performing. Keep the others running in the background on demo, not real money, to keep measuring them.
- There is no “best” algorithm; it depends on the context. Someone gambling with $10,000 they can afford to lose may trade a very different strategy from the manager of a retirement pension fund.
- Running uncorrelated strategies together is an effective way to manage drawdown and risk.
- Plan for catastrophic loss, and have redundant emergency stops and brakes to deal with it.
- Use a dynamic, mathematical position-sizing method to balance activity across all strategies in real time.
- No more than two or three rules or indicators per strategy. Keep them simple.
- Combining mediocre strategies that complement each other beats having one great strategy. Once you have one with mediocre results, develop the next to address the first one’s weaknesses: two simple strategies that fill each other’s gaps perform better together.
- Optimise portfolio allocations on net profit and drawdown, and not on maximum drawdown alone; average drawdown matters too. Measure risk with ATR over a three-month lookback to calculate volatility.
- Optimise position sizing and diversification on live incubation, not on backtesting.
- Use heat by sector to limit risk at the portfolio level.
- Diversify with a simple equal allocation to sectors, risk adjusted (DeMiguel, 2006).
- Start simple and start small. Trade micros until you have tested how the live system behaves, studied it and watched it. Don’t size up until you are comfortable with the losses you are taking.
- Determining regime change systematically (trending versus mean-reverting, bull versus bear) is too laggy to rely on. Volatility is the best filter available, though none are good. Even adaptive moving averages don’t work well.
- Most of the time, adding to a position does not improve performance. Adding an uncorrelated element is probably better than scaling in. Scaling out is fine.
- Be out of the market when it is selling off. Not for minor corrections, but reduce size on the way down, get out when losses get large enough, wait, then get back in.
- Trading is not a replacement for long-term investing. Hold a significant portion, perhaps the larger portion, of the overall portfolio in long-term buy and hold (SPY, VTSAX, or income and high-yield ETFs), and trade with a smaller portion.