Short Disclaimer

This is historical simulation research. All figures are backward-looking; they do not predict live results and you should expect slippage, differences in execution, and market evolution that can materially change outcomes.

Intro / Thesis

We tested a specific Coinbase-BTC structural idea across Kalshi's KXBTC15M binary market: rather than chase fresh 15-minute momentum, wait for a 24-hour VWAP trend to establish, then enter on a pullback into the 1-hour VWAP that begins to reclaim with short‑term momentum and 1‑minute velocity confirmation. The hypothesis is straightforward—pullbacks that respect the broader micro‑session trend produce cleaner entries than impulsive breakouts. This report summarizes 100 completed simulated variants, all built off the same core logic but with different risk, sizing, loop cadence, and entry‑filter tweaks.

Variant and Strategy Explanation

Every variant derives from a base DSL that:

• Monitors Coinbase BTC price, 24-hour VWAP, 1-hour VWAP, 5‑minute percent change, 15‑minute percent change, 1‑minute velocity, and a 12‑period EMA vs. 20‑period SMA on the 1‑minute timeframe.
• Opens a buy‑yes position when BTC is above 24-hour VWAP, below 1-hour VWAP, 5‑minute change crosses above +0.05%, 1‑minute velocity is positive, and the fast EMA is above the slow SMA—provided the contract price sits between $0.15 and $0.75, with at least 4 minutes to expiry.
• Opens a mirror buy‑no position when BTC is below 24-hour VWAP, above 1-hour VWAP, 5‑minute change is worse than –0.05%, 1‑minute velocity is negative, and the EMA is below the SMA—price band $0.25‑$0.85, same time filter.
• Exits all positions if unrealized PnL reaches +$5, max loss –$8, or expiry falls within 2 minutes.

The 100 completed variants then explore:

Each variant is saved as a fully runnable Turbine strategy; no hypothetical tweaks outside the tested set are discussed.

Top Results

Across 100 variants, absolute performance was negative. Even the top‑ranked strategies produced deeply negative returns, though the relative ranking shows which parameter combinations bled least.

The table below includes the first‑, second‑, and fifth‑ranked variants to illustrate the narrow distribution among the best:

Rank  Variant    ROI %    Sharpe    Total PnL   Trades   Win Rate
─────────────────────────────────────────────────────────────────
 1     017      –84.4    –0.01      –$21.10      238      10.6%
 2     018      –84.4    –0.01      –$21.10      238      10.6%
 5     037      –92.08    0.00      –$23.02      249      11.2%

The uppermost variants (017‑020 and 037‑040) all cluster around an 84‑92% loss with 0 to –0.01 Sharpe. The win rate settles near 10‑11% on 238‑249 trades. In each case, the strategy takes many small losers plus an occasional profit exit that covers only a fraction of the accumulated hemorrage. The PnL floor exit (–$8) triggers early and often, while the $5 profit takeoff rarely catches a move large enough to offset the frequency of small adverse fills in the KXBTC15M market.

No top variant meaningfully alters the base thesis in a profitable direction; they merely delay the drawdown slightly through more conservative entry bands and position sizing.

Bottom Results

The worst‑performing variants deepen the hole significantly and with fewer trades:

Rank  Variant    ROI %     Sharpe    Total PnL   Trades   Win Rate
──────────────────────────────────────────────────────────────────
 85    021      –259.4    –0.02      –$12.97      196      11.2%
 89    041      –259.4    –0.02      –$12.97      196      11.2%
 92    044      –259.4    –0.02      –$12.97      196      11.2%

These variants produce a –259% ROI on a smaller starting capital base. Though their total PnL in dollars (–$12.97) is less than the top cluster’s (–$21.10), the capital efficiency is dreadful: the drawdown per trade is larger, the max drawdown column (–15.88%) slightly understates the %‑of‑account damage, and the Sharpe ratio dips to –0.02. These runs typically used either looser entry filters, wider price bands, or shorter loop intervals that caused them to fire into more false reclaims, especially during sideways 1‑hour VWAP oscillations where both the bullish and bearish rules triggered in rapid succession before a meaningful trend emerged.

Conclusion

The VWAP‑reclaim pullback pattern, as specified, did not produce a positive expectancy in this simulation set. Even the best combinations of risk gates, sizing, and trigger sensitivity eroded capital at a rate that makes the strategy unviable for live deployment. The primary culprits appear to be:

1. Signal noise in the 15‑minute binary contract. BTC moving back through the 1‑hour VWAP with positive 5‑minute momentum and EMA‑cross confirmation still frequently resolves sideways or reverses within the 15‑minute window, causing repeated intra‑expiry whipsaw exits.
2. Asymmetric exit thresholds. The $5 profit exit is hit far less often than the $8 loss limit, and the 2‑minute time exit frequently crystallizes small unrealized losses instead of letting the position mature.
3. KXBTC15M contract behavior. The contract’s price tends to stick near extremes (overpriced “yes” / “no”) in strong momentum, but the pullback-reclaim pattern typically fires when contract prices are mid‑range—exactly where the payout odds are poor for small directional moves.

No variant broke the negative‑PnL ceiling, and the narrow dispersion across the top 40 or so variants suggests the core pattern itself is the limiting factor, not any particular parameter choice.

For researchers exploring similar territory, a more promising path may be to:

• Tie entry confirmation to 1‑minute bid‑ask spread behavior or actual order‑book flow, rather than pure VWAP crossover.
• Use a trailing stop or volatility‑adjusted exit instead of fixed PnL gates.
• Test on longer expiry series (e.g., KXBTC60M) where short‑term noise has more room to resolve.

As always, the full set of 100 variant DSLs is preserved and retrievable as executable Turbine strategies.

Long Disclaimer

This report is the output of automated historical simulation research and is provided for informational purposes only. It does not constitute investment advice, a recommendation to trade any security or instrument, or a forecast of future performance.

All performance numbers—total PnL, ROI, Sharpe ratio, win rate, max drawdown—are derived from pre‑recorded market data and a specified strategy DSL run in a simulated environment. They do not account for:

• Live exchange latency, price slippage, or order‑book depth.
• Regime shifts in crypto volatility or correlation structure.
• Kalshi’s specific execution logic, order‑cancellation behavior, or fee schedules that may have changed since the tested period.
• Psychological or operational friction involved in real‑time decision‑making.

Past simulation results, including any ranked variant, do not guarantee future outcomes. A negative‑performing historical strategy should not be “reversed” under the assumption that opposite PnL will materialize; structural asymmetry in binary options pricing may persist in both directions.

The strategy DSLs referenced are machine‑generated artifacts; their labeling (“Kalshi variant 017,” etc.) is internal and carries no qualitative implication. The research team may or may not run any of these strategies with live capital.

By reading further, you acknowledge that you are solely responsible for any trading decisions you make, and that you have not relied on this document as the basis for any financial commitment.