Module 9 of 14 · Building a portfolio

Combining rules and markets

You now have several forecasts per market — trend at a few speeds, carry — and many markets. Diversification is the only free lunch in finance, and there are two places to eat it.

Across rules. Combine forecasts with fixed weights (for example 60% trend spread across speeds, 40% carry). Because the rules are imperfectly correlated, the blend is less volatile than its parts and its average size falls below 10. A forecast diversification multiplier scales it back up, then you cap the result at ±20 again.

Across markets. Give each market an instrument weight, size each position as if its share of capital were the whole account, then multiply every position by an instrument diversification multiplier so the portfolio as a whole hits your risk target. Both multipliers come from the same formula, 1 ÷ √(wᵀ ρ w), using the correlation matrix of returns. Cap them, because correlations are estimated with error and rise sharply in crises.

Weights should be boring. Equal weights within a group, groups weighted by asset class, and a strong preference for leaving weights alone over chasing whatever backtested best. Optimised weights fitted to history are usually just fitted noise.

🎛 Diversification lab

8
0.25
0.3
2.5

Diversification multiplier

1.71×

positions scaled up by

Portfolio Sharpe

0.51

from 0.30 each

Ceiling as N → ∞

0.60

SR ÷ √ρ

Risk vs target

100%

full target taken

For N equally weighted components with the same Sharpe ratio and the same pairwise correlation ρ, the blend's volatility shrinks by √(N ÷ (1 + (N−1)ρ)). The diversification multiplier scales positions back up by that factor so the blend still hits its risk target, and the Sharpe ratio rises by the same factor. Notice the ceiling: as N grows, the portfolio Sharpe approaches SR ÷ √ρ and stops improving. The same arithmetic applies to blending trading rules (forecast diversification) and to blending markets (instrument diversification).

Markets beat rules

Trading rules for the same market are usually highly correlated — two trend speeds might correlate 0.8 or more — so adding a seventh trend variant buys very little. Markets from different asset classes often correlate 0.1–0.3, which is where the curve in the lab climbs steeply. Going from one market to twenty can multiply the Sharpe ratio several times; going from three rules to six rarely adds more than a little.

The binding constraint is capital, not ideas: each additional market needs enough money to hold whole contracts (Module 2). Module 13 shows how to get most of the diversification benefit even when you can't afford a position in every market.

Things to try

  • • Set ρ to 0.8 (several speeds of one rule) and move N from 2 to 10: the portfolio Sharpe barely moves.
  • • Set ρ to 0.2 (markets across asset classes) and do the same — that's the diversification curve worth paying for.
  • • Lower the multiplier cap to 1.5 with ρ = 0.1 and N = 30: the cap now binds, so you deliberately run below target risk. That is the price of not trusting low correlations.

Original interactive lesson. Labs run on a seeded synthetic market in your browser, built to illustrate mechanics, not to predict any real market. Contract figures are illustrative; check exchange specifications. Educational analysis, not investment advice.

How are we doing?