Backtests

Can combining signals reach a Sharpe ratio of 1?

In plain English: not even on the years we used to pick them. On the years that followed, no. Choosing the signals that had worked in 2010–2017 gave a combined portfolio with a Sharpe ratio of 0.63 over those years. From 2018 to 2026, the same rules managed 0.04: +0.0% a year, with much bigger swings. A fall like that is what overfitting looks like, and why we judge every strategy on years it was not chosen on.

Each signal on its own

In 2010–2017 the signals' quarterly results had correlations from -0.37 to 0.77 (average 0.14, 0.27 ignoring sign). The chosen signals (shares in issue falling, net buybacks, gross profitability) had an average correlation of 0.28 with each other, higher than the signals overall, so combining them adds less than their number suggests. 5 of the 11 signals had a lower Sharpe ratio after 2017 than before.

Can combining strategies beat the market?

In plain English: no. A strategy mix can have a higher Sharpe ratio than any of its parts only if the parts are unrelated and each still works. Every strategy here looked strong in 2012–2017, when even the S&P 500 managed a Sharpe ratio of 1.82 on quarterly returns. From 2018 the market-neutral mix returned −2.1% a year (Sharpe -0.15), and adding it to the S&P 500 gave 0.52 against 0.75 for the index alone. Nothing we tested kept a Sharpe ratio above 1 after 2017; simply holding the S&P 500 had the best Sharpe ratio over the whole period.

Quarterly returns. The market-neutral mix weights each of its four strategies by the inverse of its swing over the previous 12 quarters and is scaled to 6% a year of swing, using only past quarters. The trend rule holds the S&P 500 when it ended the previous month above its 10-month average and Treasury bills otherwise. The insider strategy is the one chosen on 2010–2017, with the market hedged out using its 2010–2017 beta. Sharpe ratio is return above Treasury bills per unit of swing (nothing subtracted for market-neutral strategies). Today's companies only; before costs. Not investment advice.

Would buying the rankings have paid?

Each quarter since 2010, the top 100 of each list as it stood then, in equal amounts, against the S&P 500 bought and held. Today's companies only (failed and acquired firms are missing, which flatters the lists), and fewer companies qualify in the early years. Size and profit alone are not an edge: none of the lists beat the market by more than chance would explain once its market risk is allowed for.