DCA and grid bots: when they work and how they kill accounts

Grid and DCA are the two most-implemented automated strategies, and the two most misunderstood. Both are mechanically simple, both have high win rates, and both have a specific way of failing that the win rate conceals.

The shared payoff shape

Both strategies take many small wins and occasionally a very large loss. In a range they look excellent; in a trend against the position they accumulate.

This shape is not a flaw — plenty of profitable strategies have it. The problem is that the win rate is a bad summary of it. A bot at 94% winning trades sounds outstanding and tells you nothing about the 6%, which is where the entire distribution lives.

DCA: the exposure ceiling is the whole design

A DCA ladder adds to a position as it moves against you, lowering the average entry. Each addition improves the break-even and increases total exposure.

Without a hard ceiling this is unbounded, and the failure is not gradual: the position that has consumed six ladder steps is the largest position you have ever held, in the instrument that has moved furthest against you, at the moment you have least free margin.

# The ceiling is the strategy. Everything else is parameters.
class DcaLadder:
    def can_add(self, position, equity, cfg) -> bool:
        if position.steps_used >= cfg.max_steps:          return False
        if position.notional >= equity * cfg.max_exposure: return False
        if position.unrealised_pct <= -cfg.abandon_at:    return False   # take the loss
        return True

That third condition is the one people leave out. A DCA ladder needs a point at which it stops averaging and accepts the loss. Without it, the strategy's implicit assumption is that price always comes back — which is true right up until the instrument that does not.

Grid: the fee arithmetic decides viability

A grid places buy and sell orders at intervals through a range and profits from oscillation. Whether it works is mostly determined before any market movement, by one comparison.

Each completed round trip earns the grid step and pays two commissions. If the step is not comfortably larger than the round-trip cost, the grid is a machine for paying fees — and it will do so with an impressive win rate the whole time.

Our grid step calculator does this check, and it is the single most useful thing to compute before deploying one. The second most useful is what happens when price leaves the range — a grid outside its range is a directional position nobody chose.

Where each genuinely fits

• Grid: liquid instruments, range-bound conditions, a step comfortably above costs, and a defined action for range exit — stop, re-centre, or close.
• DCA: instruments you would be content to hold, a fixed maximum number of steps, a total exposure ceiling, and an abandon level.
• Neither: on a trending instrument against your direction, on anything illiquid, or with capital you cannot afford to have locked in the tail case.

How to test them honestly

The standard backtest flatters both strategies because the sample period usually contains more ranging than trending. Three corrections:

1. Test specifically on the worst trend in your data, not on the full period average.
2. Report maximum exposure reached and maximum drawdown, not win rate or total return.
3. Include funding if the instrument is a perpetual — a DCA position held for weeks pays it continuously.
4. Model the range exit explicitly. A grid backtest that stops at the range boundary has tested the easy half.

Both strategies are legitimate tools. They are also the two most common ways an automated account goes to zero, and the difference is entirely in whether the ceiling was designed in from the start or added after the first bad trend.