We Backtested 10 Strategies on 129,748 Hourly Bars. Every One Lost Money.

Retail traders are told to drop to lower timeframes for more opportunities. We tested whether the opportunities are real once you charge for them. They are not, and the failure is not marginal. All data below comes from our own backtests on

Retail traders are told to drop to lower timeframes for more opportunities. We tested whether the opportunities are real once you charge for them. They are not, and the failure is not marginal.

All data below comes from our own backtests on Binance public market data, with fixed strategy specifications written before the test, no parameter tuning, conservative same-bar fills resolved against the trade, and one position per setup.

The headline result

We ran a suite of ten strategies across 129,748 hourly bars of BTCUSDT and SOLUSDT, split walk-forward: everything before 1 January 2024 as the training period, everything after as verification.

In verification, all ten strategies were negative on both assets. Profit factors between 0.50 and 0.95. Totals from minus 16 percent to minus 88 percent. Not one survivor.

Sample sizes ran from 157 to 2,547 trades, which is large enough that this is a result rather than noise.

The training period looked considerably better, with profit factors between 1.0 and 1.6, and the three best performers all collapsed in verification. That gap is the lesson: a strategy that looks fine in-sample on an hourly chart is usually measuring the regime it was fitted to.

The timeframe ladder

Running the same discipline across three resolutions produces a monotonic result:

Timeframe	Verified out-of-sample outcome
Daily	Real but modest edges, profit factors 1.2 to 3.9
4 hour	Thin to zero
1 hour	Uniformly negative

Edge is monotonic in timeframe. There is no resolution below daily at which the tested strategies got better, only resolutions at which they got worse.

Why: the toll is bigger than the edge

The mechanism is arithmetic, not mystique.

Average gross edge per trade at the 1-hour level measured between plus and minus 0.1 and 0.8 percent. A retail round trip at 0.1 percent per side costs 0.2 percent. On top of that, win rates at 1 hour fall 3 to 10 points against the same strategy run daily, because a larger share of each bar is noise.

This lines up with the academic momentum literature, where breakeven cost thresholds sit around 3 to 10 basis points. Retail crypto taker fees are roughly 20 basis points. You are trading a strategy that dies somewhere between those two numbers.

What the ETF era did to older backtests

A separate walk-forward run on daily data makes the same point about time rather than resolution. Strategies trained on 2020 to 2023 showed profit factors between 5 and 57. The same strategies verified on 2024 onward produced profit factors between 1.2 and 3.9, and several went negative outright.

Those enormous training-period numbers were regime artifacts. The survivors also turned out to be asset-specific rather than universal: BTC rewarded slow trend following, while SOL's post-2024 chop killed slow trend and rewarded quick capped breakouts instead. A strategy is not a strategy until you know which asset and which regime it belongs to.

The one short-horizon survivor

Out of a full day-trading suite, exactly one setup survived verification at retail taker fees: going long SOL for 8 hours after an extreme negative funding print. Profit factor 1.70, plus 30.7 percent, 59.3 percent win rate over 54 trades.

Two things about it matter more than the number. It is event-driven, occurring roughly twice a month and clustering in volatile weeks, so it cannot be turned into a daily routine. And its edge is concentrated in the first hours after the print: an 8-hour hold outperformed the multi-day version of the same signal.

At maker-tier fees, 0.02 percent per side, a second day trade becomes viable, a volatility breakout on BTC with a profit factor of 1.32. That is a fee-structure result, not a strategy result. The trade did not change. The toll did.

Meanwhile, raw session drift, the thing that schedule-based day-trading plans are built on, measured between 0.04 and 0.15 percent per session. All of it below the 0.2 percent round trip. Schedule-driven day trading is structurally fee-dead at retail pricing.

Method

Binance public API data, BTCUSDT from August 2017 and SOLUSDT from August 2020, daily and 4-hour series totalling roughly 38,000 bars, plus a 1-hour series of 129,748 bars. Fees charged at 0.1 percent per side for the retail case and 0.02 percent per side for the maker case. Specifications fixed before running, no tuning between runs, same-bar fills resolved as losses where ambiguous, one position per setup at a time.

Verification samples in the daily walk-forward were smaller, 7 to 49 trades in places, and we treat those as provisional and re-run them quarterly. The 1-hour result does not have that problem.

One honest caveat: this is two crypto assets, not a cross-market claim. What transfers is the method and the arithmetic. The toll is bigger than the edge is a statement about cost per trade against edge per trade, and that comparison is worth running on whatever you actually trade before you drop a timeframe.