Quant Strategy Validation
TRPZLabs Quant Radar
A tool that stress-tests a trading strategy in thousands of variations at once, then throws out the ones that only looked good by luck, so you are left with an edge that survives real markets, not a curve-fit that dies live.
- ~77×
- faster than the obvious way
- 5
- checks against luck
- 8
- timeframes
The problem
Most trading strategies are wishful thinking: they look brilliant on past charts and quietly die in live markets. Testing every variation strictly enough to catch that, across dozens of instruments, is painfully slow.
What I did
I built a bench that runs the whole set of variations in one fast pass, then puts each survivor through five independent luck checks. A strategy only passes if it beats buy-and-hold, beats random noise and clears its own confidence check. Every run is logged so any result can be reproduced.
Engineering decisions
- 01
Thousands of tests in one pass
Instead of testing each variation one by one, the whole set runs together in one pass, about 77 times faster (roughly 13 minutes down to 10 seconds per instrument). It also models leverage that the underlying library does not support, exactly.
- 02
Five ways to catch a fluke
A setup has to survive five independent checks and beat buy-and-hold, random noise and its own confidence bound. Setups that fail stay visible and flagged, so you always see how much of a result was luck.
- 03
Cross-checked against TradingView
The indicators are built to line up with TradingView, and a mirror of the engine lets you re-check a found setup right on the chart. They are not identical, but you can always look and compare, so nothing is a black box.
- 04
Warmed-up data, honest signals
Before the period under study, it loads extra history to warm up, so slow indicators like a 200-bar average are already accurate on the first bar of the test and do not throw false signals at the start.
Screens
Outcome
A fund-grade research tool built solo in plain Python: it tests strategies about 77 times faster than the obvious approach, filters out luck with five independent checks, models leverage exactly, and logs every run so results reproduce. Plus a risk desk that sizes positions and flags thin or overheated markets.
Built with
- Python
- Streamlit
- VectorBT
- pandas
- NumPy
- ccxt
- SQLite
- Pine Script