Sizing a trend sleeve with a risk-aversion sweep

A second test, not a conviction vote

Editorial: treat the capital share of a locked mechanical trend book as a second test, not as a conviction vote. After the entry and exit rules are fixed, an allocation sweep under quadratic utility shows how the trading sleeve sits beside stocks and bonds as the variance penalty changes.

That reading is a TradersWeek interpretation of the historical workflow. The archive itself records a locked rule, a spreadsheet optimizer, and the path of weights as caution rises.

The locked trading sleeve

A mechanical point-and-figure rule with 2 percent boxes and three-box reversals is applied to gold, euro-dollar, and the S&P 500. Those systems form the trading sleeve next to conventional stock and bond holdings.

The point-and-figure rule stays fixed while the capital share assigned to it is varied. The sizing step does not rewrite the entry or exit procedure.

How quadratic utility ranks mixes

Portfolio shares are chosen by maximizing expected return minus a risk-aversion coefficient times expected portfolio variance. Quadratic utility is that scoring rule: it ranks a mix by expected return reduced by a variance penalty.

A risk-aversion coefficient of zero ignores variance. Larger values treat variance as more costly when ranking mixes. A cited population average sets that coefficient at 2, with wealthier holders described as using a lower value.

What the allocation sweep shows

The allocation sweep uses observations from January 2000 through February 2007. A spreadsheet optimizer produces an allocation mix at each risk-aversion setting from zero through a very high reading of six. Repeating the same optimizer at successive settings lets the path of weights be read as a map of caution, not as a single recommended mix.

At zero risk aversion the optimizer assigns the entire mix to the gold trading system. Bond weight rises with risk aversion and peaks near a coefficient of 2.5, after which the trading-system weights increase again. On this sample the optimizer keeps a nonzero trading-system allocation at every tested risk-aversion level.

The sample window and a longer history

The mixes above belong to the January 2000 through February 2007 window used in the sweep. Extending the history is described as adding statistical power while risking the loss of regime-specific allocation results.