Adding a 200-day moving-average trend filter to each leg of a classic 60/40 portfolio (SPY 60% + AGG 40%, monthly rebalanced) reduced the maximum drawdown from −21.8% to −17.4% over the 2011–2026 evaluation window — but compound growth fell from 9.5% to 3.1% per year. The filter spent about 44% of the window in cash and produced repeated buy-high/sell-low whipsaws in 2015–16 and 2022–23. The drawdown insurance is real (it clearly helped in 2022); the price is much larger than the benefit in this configuration.

This study shows what this exact filter did on one evaluation window with one set of execution assumptions. It does not prove that all trend filters fail, and it is not investment advice.

Validation ≠ effectiveness, backtest ≠ proof

Short answer

“Add a trend filter” is one of the most common DIY improvements suggested for a 60/40 portfolio. The pitch is intuitive: when the stock market is below its 200-day average, why stay fully invested? Before trusting that intuition, it should be tested as a controlled experiment: same baseline, same window, same costs, one changed variable — and with the failure periods and cash exposure shown, not just the CAGR.

Why this question matters

A per-asset 200-day MA trend gate should cut the worst drawdowns of a 60/40 portfolio by moving a leg to cash when its own trend is down. The open question is the price: how much return, cash exposure and turnover change in exchange for the drawdown insurance.

Hypothesis

In the trend-filtered variant each leg is evaluated independently: when a leg closes below its own 200-day moving average it moves to cash at the next monthly rebalance, and returns to its target weight only when it closes back above. A leg that is out of trend does not transfer its weight to the other leg — the freed weight sits in cash. This is the engine’s documented “trend filter + cash / risk-off” semantics.

Warm-up is explicit: the data source serves US daily prices from 2010-01-01 (shared data contract), so a 200-day MA cannot be computed until late 2010. Both variants run from 2010-01-01, but every reported metric, table and chart covers only the evaluation window starting 2011-01-03, with NAV normalized to 1 at that point. This keeps the comparison free of indicator warm-up bias.

Baseline and changed variable

Exact rules

The metric grid, comparison table, annual returns and charts below are rendered directly from the study’s evidence file (real engine runs). Annual returns make the pattern clear: the filter helped in 2022 (−7.2% vs baseline −16.2%) and shrank the COVID drawdown, then gave back far more than it saved — 2017 +0.4% vs +14.4%, 2023 −9.5% vs +17.8%, 2025 +6.0% vs +14.0%.

Results and evidence

After dodging part of the 2022 bear market, the filter re-entered stocks in early 2023 and was whipsawed twice: bought SPY at $396.65 in February, sold at $365.05 in March; bought again at $436.65 in August, sold at $402.63 in October. The variant lost −9.5% in 2023 while the plain 60/40 gained +17.8% — a 27pp gap in one year.

One important failure period: 2023

Where the strategy failed

Counter-evidence — the filter did provide real insurance: it cut the 2022 calendar-year loss from −16.2% to −7.2%, reduced the COVID drawdown from −21.8% to −9.9%, and lowered volatility from 10.5% to 6.2%. Those benefits are real. The question this study answers is the price: a 6.4pp annual cost and ~44% average cash for a 4.4pp drawdown improvement.

Limitations and counter-evidence

The naive per-asset 200-day MA trend filter on a 60/40 portfolio is not a free lunch: it trades a 4.4pp drawdown improvement for a 6.4pp annual return cost, ~44% average cash, and a 2022 drawdown from which it never recovered within the evaluation window. The drawdown insurance was real — most visible in 2022 — but the whipsaw and cash-drag costs were much larger. This does not condemn all trend filters; it defines the baseline against which smarter variants (equity-only gate, T-bill fallback, slope confirmation, daily execution) should be measured.

Verdict: failed as implemented

Verdict

Adding an absolute-momentum gate — hold the relative-momentum winner only while its own 12-month return is positive, otherwise cash — to a monthly top-1 rotation over SPY/IWM/EFA did NOT reduce drawdown over the 2011–2026 evaluation window. The maximum drawdown went from −33.4% to −35.7%, while compound growth fell from 9.0% to 4.6% a year and a $100,000 stake ended at $374,833 vs $199,889. The gate spent about 12% of the window in cash and clearly helped only in 2022 (−13.3% vs −15.7%). Everywhere else the protection cost more than it saved: the 12-month lookback exits late and re-enters late, so the rotation systematically missed the 2019, 2020 and 2023 recoveries.

This study isolates one changed variable — the absolute-momentum gate — on one evaluation window with one set of execution assumptions. It does not prove that all momentum gates fail, and it is not investment advice.

Controlled comparison, not a strategy recommendation

“Dual momentum” is one of the most searched strategy families on the web. The pitch is intuitive: momentum rotation picks the strongest asset, and an absolute-momentum gate (only hold it if it is also rising) is supposed to protect against bear markets. Before trusting that intuition it should be tested as a controlled experiment: same universe, same lookback, same rebalance, same costs, one changed variable — with the failure periods, cash exposure and missed recoveries shown, not just the CAGR.

This is deliberately a “Dual Momentum-style controlled implementation”, not an exact reproduction of Gary Antonacci’s published strategy. The live Strategy DSL supports top-N 12-month relative momentum with a positive-momentum eligibility gate and a cash fallback, which is the closest faithful setup. The difference matters: the published results describe exactly this implementation, on exactly this window.

An absolute-momentum gate should cut the worst drawdowns of a momentum rotation by moving to cash when the selected asset’s own 12-month momentum turns negative. The open question is the price: how much return, time-in-cash and missed-recovery cost the protection carries, and whether the gate exits and re-enters at systematically bad times.

Both sides are the same canonical Strategy DSL graph — indicators, cadence, ranking, TopN(n=1), equal-weight target vector, RebalancingCapitalStrategy with 100% gross exposure — except for one node: the eligibility series that filters symbols before ranking. The baseline uses an always-true condition (close price > 0); the variant uses the winner’s 12-month momentum > 0. When the gate fails, no symbol is eligible, TopN selects nothing and the equal-weight vector is all zeros, so the portfolio holds cash until the next monthly decision.

Warm-up is explicit: a 252-session momentum lookback needs ~1 year of data, and the shared price source serves US daily data from 2010-01-01. Both runs were all-cash through 2010 (momentum not yet computable). Every reported metric, table and chart covers only the evaluation window starting 2011-01-03, with NAV normalized to 1 at that point.

The metric grid, comparison table, annual returns and charts below are rendered directly from the study’s evidence file (real Agent Lab runs, evaluation window only). Annual returns show the pattern: the gate’s impact is concentrated in 2011 (−24.4% vs −12.7%), 2012 (+7.8% vs +15.1%), 2019 (+10.9% vs +25.7%), 2020 (−5.2% vs +18.2%), 2022 (−13.3% vs −15.7%) and 2023 (+8.8% vs +19.0%). In 2013, 2014, 2017, 2018, 2021 and 2024–2026 the winner’s momentum never turned negative, so the gate never triggered and the two portfolios are nearly identical.

A 12-month momentum lookback turns negative months after a peak and turns positive months after a bottom. The variant sold SPY on 2022-06-02 at $386.53 (momentum −1.3% at the June decision), sat in cash for 209 sessions, then re-entered EFA on 2023-04-04 at $64.41 — after SPY had already recovered from its October 2022 low. The same pattern hurt in 2019–2020: sold SPY 2020-04-02 at $224.73 near the COVID bottom and re-bought at $280.98 on 2020-06-02, missing the entire V-shaped recovery (−5.2% vs +18.2% in 2020).

The gate exits late and re-enters late

Counter-evidence — the gate did provide some real protection. It cut the 2022 calendar-year loss from −15.7% to −13.3%, trimmed 2015 (−1.6% vs −3.4%), and lowered annualized volatility from 18.6% to 17.2%. The 2022 avoidance is genuine: the gate exited SPY in June 2022 and stayed out for most of the bear market. The problem is that over the full window the avoided losses are smaller than the missed recoveries: a 2.3pp worse maximum drawdown and a 4.4pp worse annual return.

The positive-momentum absolute gate on this 12-month top-1 rotation is not a free lunch: it made the maximum drawdown slightly worse (−33.4% → −35.7%) while cutting compound growth nearly in half (9.0% → 4.6%). The protection was real only in 2022 and 2015; the 12-month lookback’s late exits and late re-entries systematically missed the 2019, 2020 and 2023 recoveries. This does not condemn all absolute-momentum gates — it defines the baseline against which faster gates (6-month lookback), earning fallbacks (T-bills) and blended momentum (1- and 3-month) should be measured.

Holding only the top-ranked ETF in a monthly SPY/QQQ/TLT/GLD momentum rotation earned a little more than holding the top two equally weighted and rebalanced monthly — 12.3% vs 11.3% a year over 2011–2026, $593,082 vs $513,439 on $100,000 — but the extra ~1.0pp of annual return came with a far worse ride: a maximum drawdown of −41.3% vs −25.8%, and a single-position portfolio that spent about five and a half years underwater starting in February 2015. On risk-adjusted measures the Top-2 portfolio was clearly better (Sharpe 0.75 vs 0.63; Calmar 0.44 vs 0.30). Top-1’s entire advantage came from two all-in years — full QQQ through the 2020 recovery (+47%) and full GLD through the 2025 gold rally (+62%).

This study isolates one changed variable — how many ETFs the rotation holds — on one evaluation window with one set of execution assumptions. It is not investment advice.

Controlled comparison, not a recommendation

“How many positions should a momentum rotation hold?” is one of the first design decisions anyone building an ETF rotation faces. One position maximizes exposure to the strongest trend but bets everything on getting the ranking right; two positions dilute that bet but cut the damage when the top-ranked asset is a false signal. This is a selection-breadth question, tested as a controlled experiment: same universe, same lookback, same rebalance, one changed variable.

Top-1 concentration may capture more of the strongest trend and therefore improve upside in persistent regimes, but it should also create greater concentration and path risk. Top-2 should diversify single-asset timing errors and may reduce drawdowns or failure severity, at the possible cost of diluted momentum exposure.

Both sides are the same rotation rule — 12-month momentum, ranked descending, monthly rebalance — differing only in how many winners are selected: one (100%) or two (50/50, rebalanced back to 50/50 at every monthly decision). There is no absolute-momentum gate, no moving-average filter, no volatility filter and no optimizer anywhere in the design. When the top-ranked asset changes, the Top-1 portfolio sells its entire position and buys the new winner; the Top-2 portfolio swaps only the half of the portfolio whose holding dropped out of the top two, and each month trims the better-performing holding back toward 50/50.

The metric grid, comparison table, annual returns and charts below are rendered directly from the study’s evidence file (real Agent Lab runs, evaluation window only). The annual returns show where the two strategies diverge: Top 1 dominated in 2020 (+47.1% vs +23.9%) and 2025 (+62.2% vs +40.4%) by being fully in the year’s hottest asset (QQQ, then GLD); Top 2 dominated in 2015 (+0.2% vs −16.0%), 2016 (−16.9% vs −23.5%) and 2019 (+14.8% vs +3.3%) by holding a second, less whipsawed asset.

From its February 2015 peak, the single-position portfolio did not return to breakeven until July 2020 — a −41.3% drawdown that lasted about 5.5 years. The damage was concentrated in 2015–16: the top-ranked ETF switched its entire holding fourteen times between QQQ, TLT and GLD across those two years (QQQ → TLT in January 2015, TLT → QQQ in March, back and forth through 2015–16, into GLD in May 2016), with the portfolio buying highs and selling lows on nearly every switch. 2015 returned −16.0% and 2016 −23.5%. The Top-2 portfolio held a second position through the same period (QQQ+TLT in 2015, GLD+TLT in 2016), cutting 2015 to +0.2% and 2016 to −16.9%.

Top 1’s worst stretch: five and a half years underwater (Feb 2015 → Jul 2020)

Where each strategy failed — and where it helped

Counter-evidence — the concentration hypothesis has a real, measurable upside. Top-1’s extra return is not noise: it came from being 100% in the strongest asset during two powerful sustained regimes (2020 tech, 2025 gold), and over the full window it compounded to $80,000 more on $100,000. The question this study answers is whether that upside was fairly priced: at 15.5pp more maximum drawdown, worse Sharpe and Calmar, and a five-year underwater period, the risk-adjusted answer is no — but an investor who can tolerate and time the concentration risk would have been paid for it.

Top-1 did earn more (12.3% vs 11.3% CAGR, $593k vs $513k on $100k) and its two best years (2020, 2025) are genuine concentration wins. But the price was 15.5pp more maximum drawdown (−41.3% vs −25.8%), a 5.5-year underwater period, worse Sharpe (0.63 vs 0.75) and Calmar (0.30 vs 0.44), and ~70pp/yr more dollar turnover. Top-2’s diversification did not merely dilute returns — it bought real protection in 2015–16 and 2019, the whipsaw years where single-position ranking errors hurt most. On risk-adjusted terms the two-position rotation is the better trade-off; an investor choosing Top-1 is explicitly buying upside for deeper, longer drawdowns.

Research

60/40 Portfolio With a Trend Filter: What Actually Changes?

Dual Momentum Backtest: Does the Absolute-Momentum Gate Reduce Drawdown?

SPY / QQQ / TLT / GLD Momentum Rotation: Does Top-1 Concentration Beat Top-2 Diversification?