Quant Researcher for Intraday Edge Audit
Budget / Salary$250–750
TypeFreelance project
LocationRemote
Posted1 hour ago
We are a small independent team running a fully automated intraday system on Interactive Brokers (TWS API, Python). We need an experienced quantitative researcher to audit our backtest and earch for changes that make the strategy genuinely profitable. You deliver evidence and rules. We implement them.
The strategy today
Long-only, intraday, US small caps that gap up 14-55% at the open (sector, ATR and liquidity filters).
Entry around 09:32-09:35 ET on a +0.5% confirmation. Exit by a 35% trailing stop, a tighter trail when momentum stalls, a -15% hard stop, or at 15:50.
10-year backtest (2016-2026, about 330 trades, USD 200 per position, USD 0.70 commission): about +USD 450, and about +USD 200 with measured exit slippage (t about 0.5). In short, no demonstrable edge yet.
What we want
This is not a narrow audit. We want you to look for whatever changes could make this strategy profitable: new entry signals, exit logic, regime filters, candidate selection, a wider or different candidate universe, or a redesign of the logic. You may propose changes beyond the ones listed below. The only requirement is that the search is honest and the validation is strict.
Fixed constraints
IBKR CASH account, no margin, no shorting, T+1 settlement. Positions of USD 100-200.
Intraday only. Every signal must be computable from data available at decision time (1-minute bars and IBKR real-time quotes; our scanner runs at 09:31 ET).
Already tested (please do not repeat): gap band and cap, ATR floor, sector filter, liquidity floors, ranking (incl. ATR vs gap), number of positions, trailing-stop width, stall-trail timing and width, take-profit, hard-stop level, time-stops, opening-range confirmation at 09:45, entry timing, halt handling, drawdown pause rules. None beat our baseline out of sample.
What we provide
10 years of point-in-time candidate days (about 19,600 rows) with their 1-minute bars, as Parquet/CSV.
Our cost and fill model, and the baseline trade log to reproduce.
Notes on known biases (our data under-represents delisted stocks).
2026 is a blind holdout. You will not receive it.
Rules of the search (non-negotiable)
Pre-registration. Before testing anything, you write a plan: the families of ideas you will explore, the metric, the pass/fail criterion, and a maximum number of variants.
Variant log. Every variant you try is logged, including failures. The total trial count is reported.
Multiple-testing correction. Significance is adjusted for the number of variants tried (for example Holm, or a deflated Sharpe ratio). More variants means a higher bar.
Walk-forward. Parameters are chosen only on past years, never on the year being tested.
Stress tests. Cost and slippage stress, and Monte Carlo.
Pass rule for any change: its out-of-sample improvement per year has a 95% confidence interval above zero in the walk-forward years, and stays positive with 3 extra points of hard-stop slippage.
Payment
Base fee: USD 350, paid by milestone, regardless of the result. A well-documented “no edge found” earns the full base fee.
Performance bonus: USD 350 (total USD 700), paid only if your final rules pass the walk-forward criteria above and the 2026 holdout test below.
The strategy today
Long-only, intraday, US small caps that gap up 14-55% at the open (sector, ATR and liquidity filters).
Entry around 09:32-09:35 ET on a +0.5% confirmation. Exit by a 35% trailing stop, a tighter trail when momentum stalls, a -15% hard stop, or at 15:50.
10-year backtest (2016-2026, about 330 trades, USD 200 per position, USD 0.70 commission): about +USD 450, and about +USD 200 with measured exit slippage (t about 0.5). In short, no demonstrable edge yet.
What we want
This is not a narrow audit. We want you to look for whatever changes could make this strategy profitable: new entry signals, exit logic, regime filters, candidate selection, a wider or different candidate universe, or a redesign of the logic. You may propose changes beyond the ones listed below. The only requirement is that the search is honest and the validation is strict.
Fixed constraints
IBKR CASH account, no margin, no shorting, T+1 settlement. Positions of USD 100-200.
Intraday only. Every signal must be computable from data available at decision time (1-minute bars and IBKR real-time quotes; our scanner runs at 09:31 ET).
Already tested (please do not repeat): gap band and cap, ATR floor, sector filter, liquidity floors, ranking (incl. ATR vs gap), number of positions, trailing-stop width, stall-trail timing and width, take-profit, hard-stop level, time-stops, opening-range confirmation at 09:45, entry timing, halt handling, drawdown pause rules. None beat our baseline out of sample.
What we provide
10 years of point-in-time candidate days (about 19,600 rows) with their 1-minute bars, as Parquet/CSV.
Our cost and fill model, and the baseline trade log to reproduce.
Notes on known biases (our data under-represents delisted stocks).
2026 is a blind holdout. You will not receive it.
Rules of the search (non-negotiable)
Pre-registration. Before testing anything, you write a plan: the families of ideas you will explore, the metric, the pass/fail criterion, and a maximum number of variants.
Variant log. Every variant you try is logged, including failures. The total trial count is reported.
Multiple-testing correction. Significance is adjusted for the number of variants tried (for example Holm, or a deflated Sharpe ratio). More variants means a higher bar.
Walk-forward. Parameters are chosen only on past years, never on the year being tested.
Stress tests. Cost and slippage stress, and Monte Carlo.
Pass rule for any change: its out-of-sample improvement per year has a 95% confidence interval above zero in the walk-forward years, and stays positive with 3 extra points of hard-stop slippage.
Payment
Base fee: USD 350, paid by milestone, regardless of the result. A well-documented “no edge found” earns the full base fee.
Performance bonus: USD 350 (total USD 700), paid only if your final rules pass the walk-forward criteria above and the 2026 holdout test below.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.