Deep Research

RSI Pullback Controls

Fast RSI recovers above 50 inside a strong 5-minute RSI trend. Sweep sensitivity only to the submitted-buy-price clamps (price floor and price ceiling); the indicator causal value, exits, size and risk limits stay fixed. Baseline (.05,.95) is one cell of the grid.

Historical research only. Not investment advice.

Top Strategy Variants

Bottom Strategy Variants

BTC 15-Minute RSI Pullback — Buy-Price Clamp Sensitivity (Kalshi, Historical Simulation)

Disclaimer (short)

This is historical simulation research on Kalshi BTC 15-minute markets. It is not investment advice, not a live track record, and not a claim about future results. Numbers below come from a parameter sweep over past data and carry the usual caveats about overfitting and selection bias.

Intro / thesis

The idea tested here is simple: on Kalshi's KXBTC15M series (BTC, 15-minute contracts, most-liquid selection), wait for fast RSI to recover above 50 while a slower 5-minute RSI is already strong, then buy the corresponding side. Symmetrically, buy NO on a fast RSI break below 50 with the slow RSI weak. The fast signal is a crossover event; the slow RSI acts as a trend filter.

A prior pass established a version of this and the only thing swept here is where the strategy is allowed to buy: the submitted-buy-price clamps (price_floor and price_ceiling). Everything else — the indicator values that gate entries, the exit rules, size, and risk limits — was held fixed. The grid covered 10 floor values (0.05 to 0.49) by 10 ceiling values (0.50 to 0.95), 100 cells in total, all of which completed. The baseline (floor 0.05, ceiling 0.95) is one cell of the grid, not a reference line.

The strategy mechanics are worth stating because the clamps interact with them. Every 30 seconds the bot checks: an unrealized-PnL stop at −25 → sell everything; a held YES exits if 1-minute RSI falls below 45; a held NO exits if 1-minute RSI rises above 55. Entries require no open position or prior order, fewer than 3 portfolio positions, 180–600 seconds to expiry, a spread of 0.03 or narrower, ask between 0.30 and 0.70 on the side being bought, and the RSI crossover/trend condition. Size is 5 contracts, capped by a $5 per-market notional limit and one entry per market. Candle data is closed-only on Coinbase BTC-USD, refreshed every 30 seconds.

Note the ask band (0.30–0.70) already constrains entries. The sweep clamps are a second, tighter filter layered on top, and that is where the interesting behavior shows up.

Variant and strategy explanation

Each of the 100 cells is a runnable Turbine strategy — every successful variant in the sweep is saved as such, with its own strategy ID and slug. The grid varies only risk.price_floor and risk.price_ceiling. The entry rules, exits, sizing, and risk limits are identical across cells.

A meaningful feature of the cell grid: certain rows and columns are flat. For example, cells with ceiling 0.80, 0.85, 0.90 and 0.95 are identical to each other across all ten floors — the net PnL for floor 0.40 is 24.27 at every one of those four ceilings. This makes sense given the structure: the ask must be ≤0.70 to enter, so a ceiling above 0.70 shouldn't bind under the entry rule as written. The identical values are consistent with that reading, though the sweep output itself doesn't explain why 0.75 breaks the pattern and 0.80 does not — that discrepancy is unresolved.

Top results

RankFloor / CeilingTradesWin rateTotal PnLROI %Max drawdownSharpe
10.40 / 0.609459.1%28.97579.4−9.270.35
20.35 / 0.6011253.8%25.48509.6−10.020.30
30.40 / 0.8013657.8%24.27485.4−14.480.23
40.40 / 0.8513657.8%24.27485.4−14.480.23
50.40 / 0.9013657.8%24.27485.4−14.480.23
60.40 / 0.9513657.8%24.27485.4−14.480.23
70.45 / 0.607063.6%23.20464.0−9.270.38
80.40 / 0.7513055.7%21.26425.2−14.480.20

The winner is floor 0.40, ceiling 0.60: $28.97 net PnL across 94 trades, a 59% win rate, a max drawdown of about $9.27, and a Sharpe of 0.35. It is not a runaway outlier — rank 7 makes almost as much money with 70 trades and a higher Sharpe (0.38), which is arguably the more attractive cell on a risk-adjusted basis even though its raw PnL is lower. The top of the table is dominated by floors of 0.40 and 0.35 and ceilings of 0.60 (and, up to 0.95, anything at floor 0.40).

Marginal averages tell the same story: the best floor by mean PnL is 0.40 ($21.39) and the best ceiling by mean PnL is 0.60 ($16.47). The 0.60 ceiling stands well above the neighbors (0.55: $5.52; 0.65: $8.01), and 0.50 averages negative (−$1.20). That is a sharp, isolated peak rather than a broad plateau, which is a warning sign on its own.

The deflated Sharpe is 0.73. That is below the 0.95 bar used here, meaning the winner is not distinguishable from the luckiest of the sweep's skill-less trials. The permutation test (edge-feed scramble, 100 permutations) returned p = 0.069 — the timing edge beat 93% of scrambled-feed re-sweeps, but this is above the 0.05 threshold. Written plainly: these checks do not establish an edge beyond selection noise, and I'm not calling the strategy validated. The permutation design scrambles only the edge feed; it does not test the market-price-dependent conditions, so any edge from those remains untested by this design. Neighborhood degradation across the grid is 0.25, which is to say performance drops materially when you move off the best cell.

Bottom results

RankFloor / CeilingTradesWin rateTotal PnLROI %Max drawdownSharpe
1000.10 / 0.509628.9%−8.39−167.8−18.96−0.06
990.15 / 0.509230.2%−7.35−147.0−17.92−0.05
980.20 / 0.509031.0%−6.55−131.0−17.12−0.04
970.25 / 0.508831.7%−6.41−128.2−16.98−0.04
960.30 / 0.507434.3%−3.42−68.4−14.83−0.03
950.05 / 0.509232.6%−1.64−32.8−14.35−0.05
930.05 / 0.7017241.3%−1.25−25.0−26.160.06
940.10 / 0.7017241.3%−1.25−25.0−26.160.06

The worst cell is floor 0.10 / ceiling 0.50 at −$8.39, with a 29% win rate, a max drawdown near −$19, and a negative Sharpe. The entire bottom of the table is the ceiling-0.50 column, which is where the marginal average goes negative. Notice a couple of oddities that the data doesn't explain: the 0.05/0.50 cell posts a smaller loss (−1.64) than the 0.10/0.50 cell (−8.39) despite being ranked lower, and the 0.30/0.50 cell has fewer trades (74) than its neighbors without an obvious cause in the clamp values. I'm flagging those as unresolved rather than guessing at an accounting reason. The −$26.16 drawdown on the 0.05/0.70 cells is a reminder that low floors and mid ceilings produce a lot of trading (172 trades) and a lot of heat.

Conclusion

The sweep produces a clear, narrow winner: floor 0.40 / ceiling 0.60, at $28.97 net PnL with a

About this Kalshi strategy backtest

What question does RSI Pullback Controls investigate?

Fast RSI recovers above 50 inside a strong 5-minute RSI trend. Sweep sensitivity only to the submitted-buy-price clamps (price floor and price ceiling); the indicator causal value, exits, size and risk limits stay fixed. Baseline (.05,.95) is one cell of the grid.

Read the analysis alongside the individual variant results. Check which entry, exit, price-band or risk settings changed, which settings stayed fixed, and whether the report documents enough historical coverage to support its comparison. A different parameter value defines a different simulation; it does not establish a new source of live trading returns.

How many strategy variants are included?

This report lists 100 completed and 0 failed strategy variants. Failed comparisons do not count as successful tests with zero profit. Review trade counts and drawdown alongside net P&L, win rate and Sharpe, and read any coverage or execution caveats in the analysis. A result based on very few trades needs particular care, even when its return looks high.

Does a profitable Kalshi backtest predict live performance?

These are historical simulations. Live results can differ because of liquidity, fees, slippage, latency, market resolution, outages and data quality. Selecting the best result after comparing many variants can also overstate how reliably it will perform on new data. Compare nearby settings, inspect losing variants and test on a separate period before considering a live run. The report is research, not a promise of profit or investment advice.

How can I inspect or test a strategy from this report?

Use an available Open strategy link to inspect that variant. Review its rules, market, date window, fill assumptions, fees and trade history before changing parameters or rerunning it. Confirm which data and execution assumptions the selected market actually supports; a replay does not establish that every historical order could have filled live. For the workflow, read the Kalshi and prediction market backtesting guide. The Kalshi backtesting software page explains the product's supported testing workflow.

This report is generated from historical simulations. Backtests can be wrong or incomplete, and live trading can differ materially because of liquidity, fees, slippage, latency, market resolution, outages, and data quality. Do your own review before running any strategy.