BTC/USDT Grid Backtest: 365 Days, 35 Parameter Sets, and the 13.5% That Survived

Introduction

Grid trading has a marketing problem: it is usually sold as a machine that prints money in sideways markets, and almost never presented with the one number that matters — what happens across the entire parameter space, not just the cherry-picked configuration. This report fixes that. I ran a full parameter sweep on BTC/USDT over the trailing 365 days, testing 7 range widths (±10% to ±50% around spot) against 5 grid counts (20 to 200), for 35 distinct configurations, all on the same historical price series and the same capital base.

The result is more interesting than a single headline number. The best configuration — a ±40% range with 20 grids — returned 13.51% annualized. The worst — a ±10% range with 200 grids — returned 0.01%. That is a 1,351x spread between the best and worst cell of a table that most vendors would summarize as "grid bots earn 10–15% a year." Both cells are "a grid bot on BTC." Only one of them is worth running.

Equally important: the sweep says the dominant variable is not the range width that everyone obsesses over. It is the grid count, and the relationship runs in the opposite direction from the common intuition that "more grids = more profit." Across the full 7×5 matrix, the average annual ROI fell from 10.88% at 20 grids to 0.68% at 200 grids — a 16x collapse in the mean, with the entire spread narrowing as the grid count rose.

This report breaks down why, what the winning range actually covers on the price chart, what the fee math implies for live deployment, and where this simplified model is likely to flatter or penalize a real bot. Everything below is derived from the 35 real backtested cells, and where the data does not answer a question, I say so explicitly.

---

Data & methodology

The dataset

• Instrument: BTC/USDT spot.
• Window: trailing 365 days, replayed bar-by-bar from Binance historical candles.
• Realized price range: $58,625 – $110,541, a high/low ratio of 1.886x.
• Median price over the window: $77,322.
• Price at the end of the window: $77,625, i.e. +0.39% above the median.

That last pair of numbers is the single most important fact about this dataset, and it defines everything that follows. Over 365 days, BTC traveled from roughly $58.6k to $110.5k and all the way back to $77.6k. The ending price sits almost exactly on the median of the year's distribution — 36.6% of the way up the annual range, computed as (77,625 − 58,625) / (110,541 − 58,625). This was a round-trip year: enormous path length, near-zero net displacement. That is the specific regime in which grid