Research Note 01

The Grid Bot Illusion

The research question is stated directly in the note: does grid trading generate an independent trading edge, or does its profitability primarily come from market exposure, execution economics, and market-making infrastructure. Two independently published implementations were tested. Chen, Chen & Jang (2025, NTU), in "Dynamic Grid Trading: From Zero Expectation to Market Outperformance," derive analytically that under a symmetric random-walk model, with fees ignored and the position liquidated on a range breakout, a classical grid's expected profit is exactly zero before costs — a proof under stated assumptions, not a universal law. Their Dynamic Grid Trading (DGT) variant, tested on data from January 2021 to July 2024, shows BTC's IRR clearly beating buy-and-hold with a lower maximum drawdown, and a smaller IRR advantage for ETH; the authors themselves attribute this largely to the cumulative rise in crypto prices over the period, not to a market-neutral mechanism.

The second implementation, Nguyen & Bui (2026, Talyxion), "Delta-Neutral Grid Market Making with Adaptive Hedging," proposes a different economic mechanism entirely: grid quoting on spot hedged through futures, aiming for spread capture rather than directional exposure. Their backtest, run on Gate.io BTCUSDT over 52 trading days in November–December 2025, reports spread/grid capture contributing roughly 76% of total P&L, funding about 18%, and maker rebate about 6%, for a headline total return near 37% with a daily Sharpe above 3.0 and a maximum drawdown under 8% of NAV. The note flags several open problems in this result: an apparent mismatch between the stated 100 USDT capital and the notional size of the paper's own order sizing, at least two citations in the paper's related-work section that do not match their own bibliography entries, and a fill model that does not account for partial fills or queue position across roughly 14,000 completed cycles.

The fee-tier experiment

A fee-tier experiment, built on the paper's own stated 0.10% per-cycle spread, recomputes the same spot-leg cycle at three fee levels. At the paper's own assumed maker rebate of −0.025% per side, the cycle nets +0.15% per cycle, meaning the exchange pays the trader. At a standard retail maker rate with a 25% BNB discount (0.075% per side), the same cycle nets −0.05%. At the standard retail maker rate with no discount (0.10% per side), it nets −0.10%. The sign of the same mechanism flips purely on which fee tier is assumed, without any change to the strategy itself.

Verdict

The verdict treats grid trading as at least three mechanically distinct approaches rather than one strategy. The classical grid has no positive expected value under the stated random-walk model. Dynamic Grid Trading's reported outperformance likely combines genuine range-oscillation harvesting with retained directional exposure to the period's price rise. Delta-neutral grid market making is a different mechanism — spread capture — whose reported backtest carries an unresolved capital-accounting question and an incomplete fill model, making it more accurate to describe as a reported backtest result than an established property of the strategy. The overall conclusion: grid is not itself a source of edge; as an implementation grows more sophisticated, the source of any return shifts away from the grid structure and toward spread capture, fee structure, and execution quality.

Formal write-up: SSRN 7376359 ↗ (working paper, DOI 10.2139/ssrn.7376359) Full note: GitHub ↗