Williams %R on crypto: wins often, loses money

None of the 8 settings made money after costs on any bar length. With the fee alone, every setting looked spectacular on 1-minute bars; that was an artefact of how a backtest fills orders, not a profit.

How it works

Williams %R places the close within the high-low range of the last n bars on a scale from −100, at the low, to 0, at the high. Readings below −80 are called oversold. It is the stochastic oscillator turned upside down, without the smoothing.

What we tested

%R over 14 or 28 bars; buy below −80 or −90; sell above −50 or −20: 8 settings.

Every setting was run on eight bar lengths, from 1 minute to 4 hours, on Upbit's Korean-won market: 12 large coins from January 2024 to October 2026 (from January 2025 on 5- and 10-minute bars; 8 coins from April 2026 on 1- and 3-minute bars). Signals are taken at a bar's close and filled at the next bar's open, spot buying only, with equal money in each coin. Every trade pays Upbit's 0.05% fee plus one price tick of slippage. That makes 8 single-indicator settings, and 2,400 more combinations where the best of them were given a filter or an exit rule. Full method.

Results by setting and bar length

Each cell is the yearly return after costs over the whole period. An outlined cell passed the two-period test: at least +10% a year in both the first and the second period, with the second keeping at least half of the first.

Return per year after costs, %

Sea level (0) is break-even after costs: blue is under water, green to brown is land.

No setting was above water after costs on any bar length. None passed the two-period test.

Setting	1m	3m	5m	10m	15m	30m	1h	4h
Williams %R(14) buy below −80, sell above −20	−100.0	−100.0	−100.0	−99.7	−92.8	−68.4	−50.5	−46.4
Williams %R(14) buy below −80, sell above −50	−100.0	−100.0	−100.0	−99.9	−97.2	−80.4	−56.5	−45.1
Williams %R(14) buy below −90, sell above −20	−100.0	−100.0	−100.0	−99.5	−89.7	−61.5	−40.2	−39.7
Williams %R(14) buy below −90, sell above −50	−100.0	−100.0	−100.0	−99.9	−94.5	−70.2	−40.5	−36.2
Williams %R(28) buy below −80, sell above −20	−100.0	−100.0	−99.9	−94.4	−73.4	−53.6	−45.4	−33.8
Williams %R(28) buy below −80, sell above −50	−100.0	−100.0	−100.0	−97.9	−85.0	−62.6	−49.1	−35.4
Williams %R(28) buy below −90, sell above −20	−100.0	−100.0	−99.9	−92.7	−68.0	−47.4	−39.4	−31.7
Williams %R(28) buy below −90, sell above −50	−100.0	−100.0	−100.0	−96.0	−77.2	−52.3	−36.3	−31.4

Adding filters and exit rules

On each bar length, the settings with the best first-period results were combined with up to two filters (trend on a longer chart, ADX, volume, RSI range, Bitcoin's trend) or one filter and an exit rule (stop-loss, trailing stop, take-profit, time limit). Combinations were only built where this indicator ranked among the leaders in the first period.

Bars	Combinations that traded	Best that passed
1-minute	84	None passed. Best overall: −5.1% a year
3-minute	579	None passed. Best overall: −11.5% a year
5-minute	579	None passed. Best overall: 0.0% a year
10-minute	579	None passed. Best overall: +0.5% a year
15-minute	579	None passed. Best overall: +7.7% a year

What we learned

Every setting lost on every bar length after costs; on 4-hour bars the median was −35.8% a year.
On daily charts it had the highest win rate of all 34 rules in our separate test, 63.1%, and lost 24.9% over two years.
The fee-only result on 1-minute bars was a median of +53,042% a year. Fast oversold readings appear when a bar happens to close on the bid side of the order book; the backtest then buys at the bid and sells at the next open, booking the spread as profit. With one tick of slippage all of it disappeared. The one-minute trap.

Questions

Is Williams %R a good indicator for crypto?

Not for buying oversold readings, in our tests: it lost money after costs on every bar length from 1 minute to 4 hours, despite winning more than half of its trades.

Every setup with this indicator in the explorer: 4h 1h 30m 15m