Intrinsic Momentum Framed as Stop-loss/Re-entry Rules

Do asset classes generally exhibit enough price momentum to make stop-loss and re-entry rules effective for timing them? In his June 2013 paper entitled “Assessing Stop-loss and Re-entry Strategies”, Joachim Klement analyzes four stop-loss and re-entry rule pairs for six regional stock market indexes, a U.S. real estate investment trust (REIT) index, a commodity index and spot gold. Specifically, he tests:

1. Fast out-fast in (most effective when there are multiple brief corrections): Exit (re-enter) when the cumulative loss (gain) over the past 3 (3) months exceeds some specified threshold. 
2. Fast out-slow in (most effective during a downward or sideways trend): Exit (re-enter) when the cumulative loss (gain) over the past 3 (12) months exceeds some specified threshold.
3. Slow out-fast in (most effective during an upward trend with intermittent crashes): Exit (re-enter) when the cumulative loss (gain) over the past 12 (3) months exceeds some specified threshold.
4. Slow out-slow in (most effective when momentum is weak and transaction costs are high): Exit (re-enter) when the cumulative loss (gain) over the past 12 (12) months exceeds some specified threshold.

He tests ranges of stop-loss and re-entry decision thresholds. Because asset class return volatilities differ, he scales these thresholds to the annual standard deviation of returns for each asset class. He assumes a constant exit/re-entry trading friction of 0.25% and zero return on cash. For relevant tests, he defines a secular bull (bear) market as an extended subperiod of positive returns significantly above long-term average (negative or zero real returns). Using monthly asset class index returns as available during January 1970 through April 2013 in local currencies when applicable, he finds that:

• Regarding asset classes, stop-loss and re-entry rules (compared to buy-and-hold):
  • Generally suppress volatility, especially for UK stocks, emerging markets stocks and U.S. REITs.
  • Enhance absolute and risk-adjusted returns for most stock markets and U.S. REITs, but not for emerging markets stocks, commodities or gold.
• Regarding past return measurement intervals:
  • Slow out rules generally outperform fast out rules. Slow out-fast in usually beats slow out-slow in.
  • Fast out rules generally underperform buy-and-hold based on both absolute and risk-adjusted returns. In particular, high trading frictions of the fast out-fast in rule pair outweigh benefits of volatility reduction.
• Regarding stop-loss and re-entry thresholds:
  • Slow out thresholds of 0.5 to 1.5 (0.5) standard deviations generate the highest geometric (risk-adjusted) net returns. 
  • Fast in (slow in) thresholds of 0.1 to 0.25 (0.1 to 0.5) standard deviations work best. However, outcomes are less sensitive to re-entry threshold than stop-loss threshold.
• Regarding secular bull versus secular bear markets:
  • Most of the benefits of stop-loss and re-entry rule pairs accrue during bear markets. However, they are still unattractive for emerging markets stocks, gold and (usually) commodities during bear markets.
  • During bull markets, stop-loss and re-entry rule pairs are attractive only for Eurozone and Japanese stocks.
  • During bear markets, fast out-fast in beats slow out-fast in for all asset classes, with a stop-loss (re-entry) threshold of 0.1 to 0.2 (above 0.25) standard deviations generating the highest risk-adjusted returns.

In summary, evidence suggests that optimal stop-loss/re-entry past return measurement intervals and decision thresholds vary by asset class. Based on both absolute and risk-adjusted returns, the all-around best combination is slow out-fast in with stop-loss (re-entry) threshold around 0.5 (0.25) annual standard deviation.

Although framed as tests of stop-loss and re-entry rule pairs, the study resembles those addressing intrinsic momentum (see, for example, “Intrinsic Momentum Across Asset Classes”).

Cautions regarding findings include:

• Testing multiple asset classes, return measurement intervals and stop-loss/re-entry decision thresholds introduces data snooping bias, such that best-performing combinations may be lucky for the sample period. Inconsistency of results across combinations amplifies this concern.
• The study uses indexes rather than tradable assets. Trading frictions and other costs involved in constructing tradable assets from indexes would reduce asset class returns, and trading frictions may vary materially by asset class. Moreover, the assumed level of 0.25% for exit/entry trading frictions is too low for much of the sample period (see “Trading Frictions Over the Long Run”).
• As noted in the paper, decision thresholds incorporate look-ahead bias by using asset class standard deviations of returns over the entire sample period. Unreported tests based on strictly historical data mitigate this concern.
• Return on cash may be material (to the advantage of market timing) during much of the sample period.
• Results for stock indexes may be sensitive to assumptions about dividends.
• Statistical significance tests assume tame return distributions. To the extent actual distributions are wild, these tests break down.

See also “Do Stop Losses Work?” and “Using Trailing Stop Losses to Reduce Risk”.