The fee line eats the grid

A grid bot places buy and sell orders at fixed steps and collects the difference when price moves back and forth. The pitch is simple: every round trip earns one step. The arithmetic the pitch leaves out is what a round trip costs.

What one round trip pays#

• The exchange fee on the buy.
• The exchange fee on the sell.
• The spread, crossed once on the side that is not resting.
• Funding, if the inventory is a perpetual and is held across a funding time.

Add those up and you get the smallest step at which a round trip is not a loss.

def breakeven_step(fee_rate: float, spread: float, funding_per_step: float = 0.0) -> float:
    """Smallest grid step (as a fraction of price) at which one buy-sell round trip is not a loss.
    A round trip pays the fee twice, crosses the spread once and pays whatever funding
    accrues while the inventory is held."""
    return 2 * fee_rate + spread + funding_per_step

step = 0.0015                    # a hypothetical 0.15 percent grid
cost = breakeven_step(fee_rate=0.0006, spread=0.0002)
print(f"edge per round trip: {step - cost:+.4%}")   # 0.15% minus 0.14% leaves +0.0100%

These inputs are made up to show the shape. With them, a grid step that looks like 0.15 percent keeps a hundredth of a percent after costs. Nudge the fee up, or let the spread widen in a fast market, and the sign flips.

Why a backtest can lie about this#

A backtest that fills at the touched price and charges no fees will show a grid as a machine that prints one step per round trip. Put the fees and a realistic spread back in and the same grid can be a loss on every cycle. The sign of a result can change from this one line.

The other half of the risk is not in the arithmetic at all. A grid assumes price stays inside a range. When it leaves, the bot is left holding inventory on the wrong side, and no step size fixes that. That is a position limit and a stop problem, which is why a grid, like any other bot type, ends at the risk gate.

What we do with a grid request#

We ask for the exchange's real fee tier and the spread of the markets it will trade, compute the break even step, and refuse to build below it. Then we put a range limit and one stop over the whole thing.