Most writing on strategies for diversification is aimed at investors deciding how much to hold in stocks, bonds and cash. That is a real question, but it isn't the one a systematic trader asks. The trader's version is narrower: I have one rule set that seems to work. Should I run it on more markets, on more timeframes, or on neither? This post answers that from a grid we published ourselves, and the short version is that the two axes behave very differently. More markets can spread risk. More timeframes mostly re-sample the same price moves, and on fast charts they turn into fee drag.
If you came here for the difference between running several different systems, that's a separate topic, covered in our piece on why three indicators often say the same thing. Here we stay with one system and ask how far it can be stretched.
What are strategies for diversification when you trade one system?
For a portfolio investor, diversification means owning assets that don't fall together. The U.S. Securities and Exchange Commission's investor site puts it plainly: spreading money across asset categories can lower the risk of large losses, according to its guide on how to diversify your investments. FINRA makes the same point in its asset allocation and diversification primer, and adds that diversification does not guarantee a profit or protect against loss.
A systematic trader has three axes instead of asset classes. You can change the strategy type, the market, or the timeframe. A short essay on Algomatic Trading's Substack names those same three, which is the closest any ranking page gets to the trader's version of the question. It stops short of numbers, though. So let's add some.
The one-system case narrows to two levers:
- Breadth across instruments: the same rule on Gold, Silver and Bitcoin, say.
- Breadth across timeframes: the same rule on the 15-minute, 1-hour, 4-hour and daily chart of one instrument.
They look symmetrical on a spreadsheet, but they aren't.
The evidence: one trend rule, three markets, eight timeframes
In 2026 we ran a plain SuperTrend trend-following rule through a full parameter sweep and published every row. The full SuperTrend backtest results grid has 2,160 rows: 1,080 with transaction costs applied and 1,080 at zero cost. The methodology, in brief:
- ATR periods 7, 10, 14, 20 and 30, multipliers 1.0 to 5.0 in 0.5 steps, so 45 combinations.
- Eight timeframes: 1m, 3m, 5m, 15m, 30m, 1h, 4h and 1d.
- Three instruments: Gold (XAUUSD) and Silver (XAGUSD) from Dukascopy 1-minute data, 2020-01-01 to 2026-10-20, and BTCUSD from 2023-12-29, a window of about 2.8 years.
- Long/short, always in the market, no pyramiding, fills at the next bar's open, run in vectorbt with full equity reinvested each trade.
- Stated round-trip costs: 4 bps on Gold, 8 bps on Silver, 15 bps on BTC.
The zero-cost rows are a diagnostic baseline, not a result anyone can reach. Nobody trades for free, so we kept them only to measure how much of each setting's result was handed over in costs. Everything below uses the real-cost rows unless it says otherwise.
One more caveat before the numbers. BTC has a shorter history than the metals, so its rows aren't directly comparable with Gold and Silver. We'll keep it in the picture but won't pretend the three series are matched.
Does trading more timeframes diversify risk?
Mostly, no, and on fast charts it does the opposite.
Start with why: a daily SuperTrend and a 4-hour SuperTrend on Gold are both reacting to the same Gold price. When Gold trends up for three weeks, both versions go long. When it chops sideways for a month, both get whipsawed, the faster one more often. Their returns won't be identical, but their positions overlap for most of the calendar. You've bought a second opinion from someone who reads the same newspaper.
Then there's cost. Per the published grid, 391 of the 1,080 real-cost rows, or 36%, lost the entire notional under full reinvestment. Those wipeouts cluster on the 1m, 3m and 5m charts, where trade count is highest and per-trade cost compounds fastest. A further 14 zero-cost rows hit the same wall even with no costs at all. The settings guide summarises it in one line: nothing on a chart faster than 15 minutes made money after costs.
Look at the most quoted setting online, ATR 10 with multiplier 3.0. On Silver's 4-hour chart it returned about +215% over the window, and on Gold's 4-hour chart about +92%. Put the same setting on a 1-minute, 3-minute or 5-minute chart of any of the three instruments and it was wiped out every time. Same rule, same market, different speed, and the opposite result.
So adding a faster timeframe to "diversify" a working slow system often isn't diversification. It's a second bet on the same moves, made at a price that grows with every extra trade. We've seen traders treat a 5-minute copy of a daily system as a hedge. In our data it was the copy most likely to sink the account.
Is there any timeframe benefit? A small one, but only between slow charts. Two slow versions can enter and exit at different points in the same trend, which smooths the equity curve a little. That's timing variety rather than independent risk, so size it like that.
Should I run the same strategy on multiple markets?
This is where the real benefit sits, though it comes with conditions.
Different instruments do move on different drivers at least some of the time. Gold, Silver and Bitcoin trend at different moments, so a trend rule run on all three spends less of the calendar in the same drawdown than a rule run three ways on one market.
The uncomfortable part is that the top setting wasn't the same in each market. Our sweep found that the top profitable setting per instrument landed on the 4-hour chart all three times, but with different multipliers:
- Gold, 4h, ATR 30 / multiplier 2.5: total profit $153,319 (153.3%) over 2,484 days, max drawdown -25.7%, Sharpe 1.01, 340 trades.
- Silver, 4h, ATR 30 / multiplier 3.5: total profit $1,207,727 (1,207.7%) over 2,484 days, max drawdown -36.0%, Sharpe 1.50, but only 191 trades, so treat that headline with care.
- BTC, 4h, ATR 30 / multiplier 5.0: total profit $50,090 (50.1%) over 1,026 days, max drawdown -53.1%, Sharpe 0.56, 61 trades.
The breadth of what worked differed too. After costs, 95 of 360 Gold settings were profitable, 93 of 360 on Silver, and only 38 of 360 on BTC.
What does that mean for a trader who wants one system across several markets? Running one global setting everywhere is a deliberate compromise. The 2.5 multiplier that led on Gold is not the 5.0 that led on Bitcoin. If you pick one, you accept that at least two markets run a setting that wasn't their own best. Tune each one separately and you've tripled the chance that you're fitting noise, which is the overfitting problem in a new shape. Neither choice is free, so say which one you made and why.
There's also a statistical warning built into any grid this size. A Sharpe ratio that tops a big sweep is partly the search talking. That's another reason to judge each market on clusters of nearby settings that worked, not one cell.
How many markets should a trading system cover?
There's no fixed number, but there is a test each new market should pass before it's added.
- It clears its own costs on its own data. A market is not a passenger. If the rule loses money on it after realistic costs, it adds drag, not spread. In our grid, BTC had 38 profitable settings out of 360 against Gold's 95. That doesn't disqualify it, but it raises the bar.
- It adds timing independence, not just another ticker. Two equity index futures from neighbouring markets will often give the same signals within hours of each other. You've added margin and fees without adding a new source of returns.
- It holds up out of sample. Test the chosen setting on a later stretch of data it never saw, the way walk-forward testing is set up to do. A market that only works in-sample doesn't belong in the book.
- The account can carry it on a bad day. More on this below, because it's the step most people skip.
For most retail systematic traders, two to five genuinely different markets is where the effort pays. Past that, each extra instrument tends to resemble one you already hold, and the variance reduction flattens out. That's the general shape of the maths, not a figure from our sweep; measure your own set before trusting it.
Check correlation in calm and stressed periods
Before you treat two markets as separate, measure how the system's daily results move together. Don't use correlation of prices. Use correlation of the strategy's per-day P&L on each market, since that's the thing you actually hold.
Run it twice: once over a calm stretch, and once over a stressed stretch: a sharp selloff, a liquidity squeeze, a month with a big macro surprise. Markets that look independent in quiet months often converge when everyone is selling at once. Plan your sizing against the stressed number, because that's the month that decides whether the account survives. Saxo's guide on correlation makes the same general point for portfolios: relationships between assets shift over time and aren't fixed.
We're deliberately not quoting a correlation coefficient here as if we measured it. The sweep above reports per-setting results, not pairwise correlations between markets. Run the check on your own trade logs; it takes an afternoon in a spreadsheet.
Build each market's data the same way first
A multi-market backtest is only fair if every series is built on the same rules. Spot metals data and crypto perpetual data don't have rollover gaps. Futures do, and if one instrument is a back-adjusted continuous contract and another is unadjusted, the trend rule sees fake jumps on roll dates in one and not the other, and the comparison stops meaning anything.
QuantStart's article on continuous futures contracts for backtesting explains the adjustment choices and their trade-offs. We covered the TradingView side of it, including what the back-adjust button does to your chart, in our guide to TradingView continuous futures and the B-ADJ button. Sort this out before you compare a single number across markets.
Session hours matter too: Gold and Silver trade nearly round the clock on weekdays; Bitcoin trades every day of the year. A daily bar on one isn't the same slice of time as a daily bar on another. Note it, and don't over-read small differences between them.
Capital: the part correlation won't tell you
Correlation is about how results move together. Capital is about whether you can afford all of them at once. Run one system on 8 markets and you need margin in 8 places. You also need the worst simultaneous loss to fit inside the account, not just the average one.
A simple way to frame it: take each market's worst drawdown from your test, assume the bad ones land in the same month, and ask whether you'd still be trading afterwards. On our 4-hour winners alone, the drawdowns were -25.7%, -36.0% and -53.1%. Those don't add neatly, but they show why a trader running all three at full size has to plan for a bad quarter that hits more than one of them.
Position sizing is a separate decision from market selection. If you haven't picked a sizing rule yet, a tool like our Kelly criterion position-size calculator is a sensible place to see how much of the account a single bet can take before the maths turns against you.
How many uncorrelated systems do you actually need?
This is the cousin question, and it's worth a short answer because it sits on the same page of the search results. Once you have one system spread across a few markets, the next step most traders take is adding a second or third system.
The benefit of each added system shrinks fast if it's built on similar signals. Two trend-following rules with different indicators usually go long and short at nearly the same times, so they add little. The thing that quietly destroys diversification here isn't asset choice; it's signal overlap. A breakout rule, an RSI rule and a moving-average rule can all be different code reading the same price move. We dug into this in detail in confluence or correlation, so we won't repeat it. The short version: count independent ideas, not indicators.
When every market says no
Sometimes the honest output of this whole exercise is "this rule works in one place at one speed." In our grid, the strongest results all sat on the 4-hour chart, and nothing faster than 15 minutes cleared its costs. That's a narrow result, but it's still a result.
Accepting it beats the alternative. A trader who forces the rule onto six markets and four timeframes to "diversify" usually ends up with one profitable sleeve paying for five losing ones, plus a lot more trades and fees. If one market and one speed is all that passes, run that, size it carefully, and look for diversification from a genuinely different second idea later.
If you want to run the same check on your own rule, our walkthrough on how to backtest a TradingView indicator covers the setup: one instrument at a time, costs switched on, enough trades to mean something. Quantzee builds analytical tools for that kind of study. Nothing here is a recommendation to trade any market. Paper trade first, with costs on, before real money touches it.
