Short Disclaimer

This report is for informational and educational purposes only. It describes historical simulations, not live trading results. Past simulated performance does not guarantee future outcomes. Trading binary options involves risk of total loss.

Intro / Thesis

We tested whether adding a simple anti-chop momentum filter to a BTC VWAP directional breakout strategy on Kalshi 15-minute binary markets would reduce trade frequency and peak drawdowns by skipping weak crosses that immediately reverse.

The base idea: When BTC price crosses above or below its 1-hour VWAP, take a directional Yes or No position on the next 15-minute binary, but only if the cross is backed by a meaningful 5-minute price change and a confirming 1-minute velocity direction. Without filters, VWAP signals in choppy sideways markets tend to produce rapid-fire entries that immediately turn into losers.

We ran 100 variant simulations to measure whether the thesis held, how different parameterizations affected outcomes, and whether the filters meaningfully changed behavior relative to unfiltered entry.

Variant and Strategy Explanation

Core Strategy (Base DSL)

• Market: Kalshi KXBTC15M — 15-minute BTC above/below binaries.
• Risk limits: Max position $20, price floor 0.10, price ceiling 0.90, max loss $100.
• Edge data: Coinbase BTC-USD continuous feed providing price, 1-hour VWAP, 5-minute percentage change (change_5m), and 1-minute velocity (velocity_1m).
• Entry logic:
  • Buy Yes: Price > 1h VWAP, 5m change > +0.05%, 1m velocity > 0, contract price between $0.10‑$0.90, and >2 minutes to expiry.
  • Buy No: Price < 1h VWAP, 5m change < –0.05%, 1m velocity < 0, same price and time constraints.
  • Position size: 1 contract per signal.
• Exit logic:
  • Exit Yes: Price drops back below VWAP (position closed).
  • Exit No: Price rises back above VWAP (position closed).
• Loop interval: Check conditions every 10 seconds.

What the variants changed

The 100 variants explored different combinations of:

• Entry threshold magnitudes (variations on the 0.05% and velocity zero-crossings).
• Slightly different position sizing (all within small limits).
• Minor timing or condition-ordering differences that Turbine's optimizer permutes automatically.

Importantly, every variant retained the same core architecture: VWAP direction plus two momentum checks. No variant removed filters entirely (that would be a different baseline test). The question was: within the filtered approach, which specific thresholds and permutations perform best or worst?

Each successful variant is saved as a runnable Turbine strategy under its unique slug (e.g., custom-on-kxbtc15m-d9e9caaa5137).

Top Results

The top-performing variants share remarkably similar characteristics — and none are profitable. Here are the key metrics from the highest-ranked simulations:

What stands out:

• Trade count is high: 3,350–3,416 trades across the simulation period. Despite the momentum filter, the strategy still fires frequently.
• Win rates are abysmal: Barely above 2%. The strategy loses on roughly 98% of its positions.
• Max drawdowns hit the account limit: Drawdowns of –$99 to –$101 effectively wipe out the $100 risk budget. This is essentially total account destruction.
• Sharpe ratios deeply negative: –1.83 to –1.88 indicates the strategy reliably destroys value with every unit of risk taken.
• The top "ranked" variants are simply the ones that lost slightly less. Among a field of uniformly losing strategies, rank reflects marginal differences in timing that slightly delayed the inevitable drawdown.

The momentum filter did not save the VWAP directional approach. It simply produced many small losing trades that steadily burned capital.

Bottom Results

The worst-performing variants show even more aggressive failure, with higher trade counts and faster capital destruction:

Key observations:

• Trade count is slightly higher: 3,490–3,514 trades vs. 3,350 for the top group. The worst variants entered more often and died faster.
• ROI exceeds –2,000%: These variants didn't just lose money — they lost more than the account limit due to how Kalshi position sizing and sequential losses compound against a small account.
• Sharpe drops to zero: Not negative, but zero, which in context is an artifact of the calculation breaking down when losses are so extreme and consistent that variance collapses.
• Win rates are nearly identical to the top group: ~2.1%. The difference between "best" and "worst" is not about picking better directions — it's about taking slightly fewer bites at the poison apple.
• Drawdowns exceed $100: The $100 max loss limit in the DSL was breached because overlapping positions and rapid sequential entries caused losses to compound beyond the stop.

Conclusion

The anti-chop momentum filter did not salvage BTC VWAP directional trading on Kalshi 15-minute binaries. Across all 100 variants, the strategy consistently destroyed capital, with win rates barely above 2%, Sharpe ratios deeply negative or collapsing to zero, and max drawdowns exceeding the account's risk limit.

The core problem appears to be that:

1. VWAP cross signals in 15-minute windows lack edge. Even with momentum confirmation, the directional move often reverses or fails to reach the binary strike within the expiry window.
2. Binary option pricing and time decay work against the strategy. Even when direction is briefly correct, the contract price may not move enough to exit profitably before expiry or before a VWAP re-cross triggers a loss.
3. Trade frequency remains high. The filters reduced some noise, but not enough — 3,350+ trades in a simulation period is still far too many for a strategy with a sub-3% win rate.
4. Position sizing compounds losses. With sequential losing trades, the account burns down systematically.

For context, this is purely a historical simulation finding. It does not mean VWAP-based approaches are useless in all markets or timeframes. But for Kalshi's 15-minute BTC binary structure with the specified entry/exit rules, the evidence is overwhelmingly negative. A different market structure — longer expiries, different underlying, or a fundamentally different exit mechanism — would be required for any variant of this idea to have a chance.

Long Disclaimer

This research report is produced by Turbine's automated strategy simulation and analysis system. All results presented are historical simulations only, generated by replaying market data through the specified strategy logic. No real money was traded, and no live market impact was modeled.

Key limitations of this research:

• No forward-looking claims: Nothing in this report should be interpreted as a prediction of future performance, a recommendation to trade, or an endorsement of any strategy variant.
• Simulation vs. reality gap: Simulations cannot fully account for real-world factors including exchange latency, order book depth, slippage, API downtime, regulatory changes, or black-swan events that affect execution quality.
• Survivorship and overfitting risk: With 100 variants tested, the "best" results may reflect overfitting to the specific historical data used. There is no guarantee these results would replicate out-of-sample.
• Market regime dependence: The test period reflects one specific market regime. Different volatility environments, trend strengths, or correlation structures could produce entirely different outcomes.
• Binary option risks: Kalshi binary options have all-or