Oct 18, 2026
Equity-Curve Trading and Other Timing Rules: What the Evidence Says
Once you have more than a handful of strategies, the question moves from which strategy to build to when to run each one. Traders have a long list of rules for it: switch a strategy off when its equity curve drops below a moving average, cut size after a drawdown, trade only in the right market regime, only in certain hours, never on news days, or pick a fresh subset from a library every month.
I went through what well-known practitioners do and what the published tests show, for each of these rules. None of them has out-of-sample, net-of-cost support for intraday strategies like the ones StrategyQuant X builds. “Out of sample” means tested on data that wasn’t used to build or choose the rule. “Net of costs” means after spread, commission and slippage. A few narrower ideas do hold up, and they’re at the end.
In the interest of full disclosure: the StrategyQuant links in this article are affiliate links. If you start there through them I may earn a commission, at no extra cost to you.
This is a review of other people’s tests and accounts, not a backtest of my own. Where a number comes from one hand-picked example, or a source the review couldn’t fully confirm, I say so next to it.
Two camps
Practitioners split cleanly. Some switch strategies on and off by their own results or pick a monthly subset from a library: Unger Academy, Marsten Parker, Stefano Serafini, and Kevin Davey for his library. Others run everything and judge only the whole book: Davey on any single market, Laurens Bensdorp, Nick Radge, and Brendan Anderson, who runs more than 30 StrategyQuant M30 strategies on US, German and Japanese index CFDs.
Most of the evidence below sides with the second camp.
1. Switching a strategy off by its own equity curve
An equity-curve filter watches a strategy’s running P/L. When the curve falls below a moving average or past a drawdown limit, real trading stops, and a resume rule restarts it later. While it’s off, you track “shadow equity”, the paper P/L it would have made, because the resume rule needs it.
What practitioners do
| Who | Rule |
|---|---|
| Kevin Davey | Control chart of trade results. Stops on any of four signals: one trade beyond 3 standard deviations, 2 of 3 beyond 2, 4 of 5 beyond 1, or 8 in a row on one side of the mean. Second kill at twice the out-of-sample max drawdown |
| Unger Academy | Pause after 6 months without a new equity high, reactivate at the first new high |
| Marsten Parker | Off when the equity curve is below its 300-day moving average. Whole portfolio stops at −10% from inception or −15% from a later peak |
| Mike Bryant (2005) | 30-trade moving average of equity |
| Brendan Anderson | No per-strategy cutoff at all. Judges the whole portfolio against Monte Carlo drawdown ranges |
Anderson’s warning sign is three 10% drawdowns in six months when his simulation expects one or two a year. He writes about it in his StrategyQuant article on strategies that survive live markets.
The positive examples are thin. Unger Academy’s case is one RBOB 15-minute strategy, out of sample from January 2018, with costs not stated:
| Version | Profit | Max drawdown |
|---|---|---|
| Always on | $100k | about $32k |
| Pause and reactivate | $80k | $20k |
| Discard after 6 months | $55k | about $20k |
So even the case chosen to show the rule working made less money than leaving the strategy alone. It traded $20k of profit for $12k less drawdown.
Bryant’s 2005 test on one intraday mini-Russell system is more interesting. Skipping trades while equity was below its 30-trade average cut trades from 513 to 365 and max drawdown from 30% to 17%, at what he called nearly the same net profit. Cutting size by 20% below the average and raising it 20% above did better: return went from 625% to 1036% with no worse drawdown. There was no in-sample/out-of-sample split and no costs, but the shape of the idea matters. Adjusting size beat switching off.
Parker also names the cost of killing. He shelved a breakout strategy over 2010-2019, when it made 3% a year with a 17% max drawdown. It then did well in 2020.
What the tests show
The direct tests are neutral to negative.
- Cesar Alvarez, on a ConnorsRSI stock strategy over 2001-2017, found a 200-day filter on equity cut annual return by 29%, and a 12-month rate-of-change filter cut it by 15% at about the same drawdown. His verdict: “Clearly using equity curve trading does not improve your strategy.”
- StrategyQuant tested a 20-period equity moving average and found it “doesn’t improve anything” and almost always lowers profit. The write-up is on their blog.
- The largest test, from Allocate Smartly, covers more than 100 tactical allocation strategies since 1973. Monthly switching on each strategy’s own curve nearly always cut drawdown and never beat the average strategy’s Sharpe ratio, and that’s before the cost of switching.
- Davey ran a 10-bar equity filter on 5 live strategies over their real out-of-sample periods. Two did clearly better always on, one was about the same, and two made more always on but had lower drawdown with the filter. The review marks this one as disputed.
The reason is persistence. A filter that switches off after losses only pays if bad runs tend to continue. Robert Carver’s simulations put the bar at about 0.2 autocorrelation. Measured persistence is lower: across 65 US equity factors, mean monthly autocorrelation is 0.11.
Normal drawdowns can also run for years. Over a 10-year horizon, a strategy with a true Sharpe ratio of 0.5 has a 5% chance of a drawdown lasting seven years or more (Rej, Seager and Bouchaud, 2017). A formal monitor doesn’t fix this either. The standard CUSUM setup for watching fund managers gives a false alarm about every 84 months and takes about 41 months to detect a real loss of skill. I wrote about drawing that line in advance in knowing when a strategy has stopped working.
A monitor can also fire for the wrong reason. Live results ran 2-3 percentage points below backtests in Liu (2026), and Anderson reports a 40-50% Sharpe loss from in-sample to out-of-sample for randomly generated StrategyQuant strategies built without robustness tests, from an experiment he doesn’t link. Compare live results against the raw backtest and the monitor fires on that ordinary decay.
2. Cutting the whole book after a drawdown
The book-level version cuts size across all strategies as the account falls from its high.
The Turtles cut the notional account by 20% for each 10% loss of the original account, per Curtis Faith. Carver uses what he calls half compounding: losses cut the capital he sizes from, and gains rebuild it only to the original level. In his one-market Eurodollar backtest it gave 22.7% a year against 33.4% with fixed capital, volatility of 43% against 62%, and a Sharpe ratio of 0.51 against 0.54. He picked that market because it performs especially well, and he says plainly that with any profitable strategy you’ll end up with less money than with fixed capital.
Volatility targeting, the other book-level tool, mostly disappoints:
| Test | Result |
|---|---|
| Carver, whole futures book scaled to constant volatility (backtest) | Sharpe 0.93 → 0.64 |
| Cederburg et al. (2020), 103 equity strategies, out of sample | Vol-managed versions won 53 to 50, only 8 significantly. Real-time combinations did worse in 72 of 103 |
| Bongaerts, Kang and van Dijk (2020), 10 equity-index futures, real time, net of 3-5 bp | Conventional monthly targeting added +0.04 Sharpe on average. Scaling only when volatility was in its top or bottom fifth improved Sharpe in all 10 markets (+0.07 average) and cut max drawdown by 6.6 points |
Recovery is slower, too. The review’s own arithmetic, assuming a constant edge and no costs, puts recovery from a 20% loss at about 1.12x as long with half compounding, about 1.28x with a “cut twice as fast as the drawdown” rule, and 1.25x with a Turtle-style cut.
The alternative is to keep standing risk low from the start. Ed Thorp’s numbers: at full Kelly the chance of ever losing half your capital is 1/2, and at half Kelly it’s 1/8, while keeping three quarters of the growth.
3. Regime filters
A regime filter allows trading only in a chosen market state, such as trending, low volatility, or above a long moving average.
Davey tested trend filters on 8 base strategies across 40 futures markets over 2009-2025, net of costs, and they improved profit and drawdown for each one. But the study is fully in sample, and the best filter cut one strategy from 1,153 trades to 18. He calls regime switching “very hard” himself. Carver tested adjusting for structural breaks in a rule’s own performance, and it didn’t help out of sample in 7 of 8 setups.
Regime filters carry two further problems.
- Every filter you add is another search. Robert Novy-Marx showed that picking the best k of n signals is about as biased as picking the best of n^k. By the same logic, adding a filter to a strategy already chosen from hundreds of generated candidates compounds the selection. I covered that bias in my piece on data-mining bias in technical analysis.
- Volatility can work the opposite way for intraday strategies. In Zarattini, Aziz and Barbon’s SPY intraday momentum system, the Sharpe ratio rose with the VIX (in sample). A “stand aside when it’s volatile” rule can cut the best days.
If you want to detect regime change as information rather than as a switch, the statistical tools are in this post on regime detection.
4. Time-of-day filters
Restricting a strategy to certain hours or weekdays looks harmless. The evidence says it’s mostly curve fitting, unless the window was chosen for a reason before testing.
- Davey optimised a best weekday on 2009-2017 data. It beat his pre-chosen day “dramatically” in sample and lost to it “dramatically” in 2018-2019.
- Stephan Schulmeister tested 2,580 rule models on 30-minute S&P 500 futures bars. They earned 7.2% a year gross and 2.6% net. Picking the best 25 on three years and trading them for the next three earned 21% in 1986-88, −0.97% in 2001-03 and −3.81% in 2004-07.
- None of 7,846 intraday rules on US equities survived a data-snooping correction (Marshall, Cahan and Cahan, 2008).
- A 2015-2026 replication of opening-range breakouts on DAX, NQ, SPX, Dow and FTSE reproduced the gross edge, with the DAX at +0.116R per trade. After 2.5 points of round-trip cost the DAX was at −0.038R. No market beat zero net, and four of five were negative.
The hour matters most for cost. These are IG’s published index CFD spreads:
| Instrument | Cash hours | Other hours |
|---|---|---|
| Germany 40 | 1.2 points | 2 next to cash hours, 4-5 overnight |
| US 500 | 0.4 | 0.6-1.5 |
| Japan 225 | 7 | 15-30 |
A strategy that trades the DAX overnight pays three to four times the cash-hours spread. A backtest with one fixed spread understates the cost of every trade outside the main session.
5. News-day blackouts
Prop firms block or cap trading around news. FTMO bars opening or closing trades from 2 minutes before to 2 minutes after selected releases on its funded Standard accounts. Neither FTMO nor Topstep presents its rule as tested.
On returns, the case for avoiding news is weak:
- The famous pre-FOMC drift, +49 basis points in the 24 hours before a Fed decision over 1994-2011, essentially disappeared after 2015.
- Only 3 of 188 calendar effects in S&P, DAX and Nikkei futures held in both halves of 1991-2008 at 0.05% costs, and none was a DAX effect.
- No practitioner or paper has published an out-of-sample, net-of-cost test of a news blackout on intraday systems.
News windows matter for execution instead. On exchanges, spreads widen about 20 seconds before and 30 seconds after a release, and most of the adjustment is done within 4 minutes (Frino and Hill, 2001). On an H1 or M30 chart, the bar that contains the release holds most of the move, and a stop inside it can fill well past its level. Guaranteed stops don’t solve it for short-term strategies: IG’s Germany 40 guaranteed stops must sit at least 2% away.
StrategyQuant’s MT4 backtests fill stops at the intended price through gaps unless “Realistic gaps handling” is switched on. It’s in the trading options documentation. Without it, a backtest can’t tell you what news costs your stops.
6. Picking strategies from a library on recent results
The last rule is rotation. Davey ranks a library of about 200 strategies each month on 6-12 months of results and runs 20-30 of them, but the account of it offers no evidence that rotation beats a fixed set. Unger Academy’s Titan picks monthly from more than 300 systems, and Serafini trades 5-6 of about 100 based on the past year.
The independent tests don’t support it:
- Across 41 equity indices, including the DAX, Nikkei and S&P 500, and 6,406 technical rules, picking the best rule on 36 months and trading it for the next 36 earned an excess Sharpe of about zero before costs. After 25-50 basis points of costs it was significantly negative (Rink, 2023).
- 355 published strategies lost 33% (mean) to 44% (median) of their Sharpe ratio after publication. An equal-weight portfolio of all of them kept about four fifths (Quantpedia, 2023).
- A few months can’t separate a dead strategy from an unlucky one. At a daily Sharpe of about 0.03, roughly 1 in 13 months of a strategy with zero edge looks like a top-decile month (Sinclair, 2025).
Anderson’s approach is the alternative. Every 3-4 months he reruns the whole databank through the same robustness filters, treats the new months as extra out-of-sample data, keeps whatever survives, and never cuts a strategy for being in drawdown.
What does hold up
Put together, the evidence supports a narrower set of rules than the ones traders usually reach for:
| Popular rule | What the tests support instead |
|---|---|
| Switch a strategy off below its equity moving average | Change size gradually, if at all (Bryant’s ±20% beat skipping trades) |
| Cut the whole book on a drawdown ladder | Keep standing risk low and fixed from the start |
| Per-strategy regime filters | None fitted per strategy. Count any filter the builder searched as extra trials |
| Optimised trading hours | Hours chosen for a reason, and spreads priced by the hour you actually trade |
| Flat on news days | Treat the release window as gap and execution risk, not a return signal |
| Monthly rotation on recent P/L | Run the whole validated pool at equal risk and re-validate it on new data |
The tests behind this come from daily futures, US equities, factors, funds and single systems. Some are intraday, but none was run on a portfolio of data-mined intraday CFD strategies, so the strongest claim is that the popular timing rules haven’t earned a place there yet.
If you want to run these checks on your own strategies, the StrategyQuant pricing page lists what each licence includes. The custom indicators, databank columns and Monte Carlo tests I’ve published are free in the StrategyQuant Codebase.
Frequently asked questions
Does equity-curve trading work?
Not as an on/off switch, on the evidence so far. Alvarez found a 200-day filter on a stock strategy's equity cut annual return by 29%. StrategyQuant's own test of a 20-period equity moving average found it almost always lowered profit. Across more than 100 allocation strategies since 1973, switching on each strategy's own curve nearly always cut drawdown but never beat the average Sharpe ratio. The filter can only pay if good and bad runs persist, and measured persistence is below what it needs.
Should I cut position size after a drawdown?
It buys a smaller drawdown at the price of less money and a slower recovery. Carver's half-compounding rule, tested on one market, gave 22.7% a year against 33.4% with fixed capital, at almost the same Sharpe ratio. Keeping standing risk low from the start, for example at half Kelly or less, does most of the same job without a ladder.
Should I stop trading on news days?
Nobody has published an out-of-sample, after-cost test of a news blackout on intraday systems. The real risk around a release is execution: exchange spreads widen for about 20 seconds before and 30 seconds after, and a stop inside the release bar can fill past its level. A short block on new entries is defensible as risk control. Flattening the whole day has no support.
Is it better to rotate strategies monthly or run all of them?
Independent tests favour running the whole validated pool at equal risk. Across 41 equity indices and 6,406 rules, picking the best rule on 36 months and trading it the next 36 earned about nothing before costs and lost after them. 355 published strategies lost 33% to 44% of their Sharpe after publication, while an equal-weight portfolio of all of them kept about four fifths.