Stacking a relative-strength-index forecast, a trend filter, and long-only momentum

Three separately testable layers

The volatility-regime problem can be taught as three separately testable layers rather than as a single switch. The first layer is a relative-strength-index: a forecast that compares an ordered implied-volatility series with a fixed lookback average of its own daily lows and classifies the series as below or above that baseline. The second layer is a trend-filter: a persistence gate that counts how many sessions the series stays on one side of the lookback average before any order is allowed. The third layer is a momentum-strategy: a single long-only procedure that turns the filtered forecast into next-session entries, exits, or cash on two equity-proxy funds.

The momentum-strategy may act only after both quantitative layers agree, and only under next-session-execution.

A relative-strength-index forecast of implied volatility

The implied-volatility-index used as the forecast input was described as a weighted blend of out-of-the-money option prices from the first and second nearest months and as a 30-day implied-volatility reading on the broad equity market.

A chart of that series from 2003 through October 2012 was described as mostly flat to down and interrupted by spikes that arrived during equity-index selloffs. That description was offered as a negative-correlation observation.

The relative-strength-index forecast carried into the later system was a 50-session moving-average comparison on the implied-volatility series, chosen as the simple rule to keep after earlier rule tests. Calculations used the daily low of the volatility series.

A persistence trend filter that can force abstention

Two trend-filter designs were considered for cutting moving-average whipsaws. One placed standard-deviation bands around the average. The other used a consecutive-session count that is easy to inspect on a chart and to code.

The persistence-count is that consecutive-session design: an integer shorter than the moving-average lookback that tallies consecutive sessions below or above that average. The count could be any integer below 50. A count of 11 was selected because it appeared reasonable and workable in sample, and a count of 20 was also tried.

The unfiltered 50-session crossover produced about 200 trades in a three-year window and excessive whipsaws. Requiring nine consecutive sessions above the average cut activity to 18 trades over five years. A whipsaw is repeated crossings of a moving average that spawn many trades with little net displacement.

One long-only momentum procedure

The procedure being adapted went long or short whenever the volatility series sat above its 50-session average and omitted trading costs. The reconstructed momentum-strategy dropped short sales and required costs to be counted. Under the long-only-constraint, the procedure is either fully invested in the two funds or in cash.

A buy required that series to sit below its 50-session average on each of 11 lookback days, after which equal amounts of an equity-index fund and a leveraged industrial fund were bought the next session. Exit logic reused the same relative-strength-index average and trend-filter count, except the volatility series had to remain above the average for the count window before a next-session sale.

Next-session-execution is the rule that a completed signal on the volatility series is filled only on the following trading day in the equity proxies.

What the stacked procedure is meant to constrain

A facing promotional page listed three recurring execution errors a trend-aware momentum procedure is meant to constrain: holding a losing direction too long, exiting a continuing trend too early, and calling a reversal while the series keeps its path.

Editorial reading: those errors are why the layers stay separate. The relative-strength-index names whether the implied-volatility series is below or above its lookback average. The trend-filter can force abstention until the persistence-count is complete. The momentum-strategy then buys, sells, or stays in cash on the next session, and only after both quantitative layers agree.