Why Your Backtest Lies to You: Generic Data vs. Broker Data
Most backtesting tools give you a clean equity curve and a confident Sharpe ratio. Then you go live, and the strategy behaves differently than the backtest promised. The usual explanations — overfitting, bad luck, market regime change — are sometimes true. But there's a simpler, less discussed cause: you probably didn't backtest on the data you actually trade on.
The aggregation problem
Most backtesting platforms run on aggregated market data — a blended feed pulled from multiple sources and normalized into a single price series. It's convenient, cheap to license, and good enough for a rough sanity check. But it isn't what your broker actually quoted you.
Every broker fills orders against its own liquidity, with its own spread, at its own moment in time. Two brokers can show meaningfully different prices for the same instrument at the same second — different spreads, different execution latency, sometimes a different underlying feed entirely. A strategy tuned tightly around entry timing or spread-sensitive logic can look profitable on an aggregated feed and behave completely differently once it's tested against what your broker would have actually filled.
This isn't a rare edge case. It's systematic. Any strategy with a short holding period, spread-sensitive entries, or thin-liquidity instruments is exposed to it — and most retail algo strategies fall into at least one of those categories.
A concrete example
Take a DAX index strategy with a 3-pip average spread assumption baked into the backtest. If your broker's actual average spread on that instrument runs closer to 4-5 pips during your trading hours — which is common outside of peak liquidity — every single trade in your backtest was priced too optimistically. Multiply that across hundreds of trades, and a profit factor of 1.4 in the backtest can quietly become breakeven or worse in live trading. Nothing about the strategy logic changed. Only the assumption about what price you actually got did.
Why this gets overlooked
Most traders check for overfitting. Fewer check whether the price series itself reflects reality. It's an easy thing to skip, because the backtest report looks complete either way — same equity curve shape, same metrics, same confidence. The data source is invisible in the output.
How partiqon handles this differently
partiqon is built around Bring Your Own Data (BYOD): you convert your own historical price data — including tick-level bid/ask data if your broker provides it — into BAR6, partiqon's open, documented format, using DataConvert. The conversion runs locally; your raw data never leaves your machine. The backtest then runs against exactly what your broker would have quoted, not an approximation.
This doesn't make a bad strategy good. It makes the backtest result mean what you think it means: a reasonable estimate of what would have happened, rather than a number computed against market conditions you never actually traded in.
Before you optimize a strategy, question the data feed first. It's the assumption most backtests never state out loud.
Test your strategy on your own broker data. Join the waitlist →