Crash 500 and Boom 500 are structurally untradeable — 2.6 million ticks of proof

Deriv's Crash/Boom indices are among the most popular instruments in retail trading communities, usually traded one way: ride the smooth "grind" and dodge the occasional violent spike. We recorded ~2.62 million ticks per symbol over 30.4 days (one tick per second), measured everything, and can now say with numbers what many traders learn with deposits: there is no exploitable directional, stop-based, or timing edge in these instruments. Here's the anatomy of why.

1 · The bait: the cleanest trend you'll ever see

Boom 500 grinds down and spikes up; Crash 500 grinds up and spikes down. The grind is almost unbelievably clean: 97.4% of Boom's ticks move in the grind direction (Crash: 95.5%), and 100% of inter-spike runs net in the grind direction, with path efficiency of essentially 1.0 — near-zero chop. To the eye, it's free money. That's the bait.

2 · The hook: the grind exactly funds the spike

Decompose all tick movement over the month into grind-direction and spike-direction totals and they match almost perfectly — Boom: 27,967 grind vs 27,977 spike (fair to within 0.018% of total motion). The slow drip you collect riding the grind is the same money the spike takes back, all at once. Crash drifted −6% over our particular month — but that's just 0.57% of its total motion, within ~2σ of a fair heavy-tailed random walk. Neither instrument has a reliable drift edge.

3 · The kill shot: your stop is a fiction

A spike is not a fast move you can react to. It is a single tick (98–99% of spikes complete in one tick, ~480× the size of a normal tick — Boom's median spike is 4.82 price units against a 0.010 median tick). There is no liquidity between the pre-spike price and the post-spike price. A stop placed "safely beyond the spike zone" doesn't get filled at the stop. It gets filled at the far side of the gap.

We tested an honest, structurally-reasoned grind strategy (enter with the grind after a spike, modest target, hard stop beyond the recent spike extreme) over the full month at real measured spreads, under two fill models:

Read that middle row again. The wins were identical under both models — the grind profits are real. The entire difference is what a loss costs when the spike gaps your stop: a capped −1R in the fantasy becomes an average −7.4R in reality. The strategy's equity curve under naive fills was a near-straight line up. Under honest fills it loses on both instruments. This single modeling fix — stops on spike bars fill at the adverse extreme — is now permanently wired into our simulator, and it is the reason you should distrust every Crash/Boom backtest you've ever seen with a smooth curve.

4 · The escape hatch that isn't: timing

"Fine — so dodge the spikes." Spikes average one per ~502 ticks, so traders count ticks and feel safer just after a spike. The data is unambiguous: spike arrival is memoryless. The interval distribution is geometric (coefficient of variation 0.992 Boom / 0.980 Crash — a clock would be near 0); the per-tick spike probability is flat at ~1/500 no matter how long it's been (hazard 0.93–1.10× the mean across all elapsed times); intervals are uncorrelated (|r| < 0.025); and cumulative grind distance adds no signal. A spike is exactly as likely 5 ticks after the last one as 600 ticks after. The "~every 500 ticks" figure is a mean, not a schedule — the median gap is ~353 ticks and a quarter of gaps exceed 700.

5 · What this means

The instruments are fair (grind movement ≈ spike movement), inseparable (you cannot harvest the grind without holding spike risk), un-stoppable (single-tick gaps blow through stops), and un-timeable (memoryless arrivals). Every conditioning signal a trader would naturally reach for — elapsed ticks, prior interval, cumulative grind, direction — is flat. The popular "grind Crash/Boom" strategies look profitable because losses are mentally priced at the stop level; price the gap honestly and the rare spike more than erases the steady gains. The door is closed.

Honest scope note: this rules out the natural, observable conditioners. We can't prove no exotic hidden signal exists — only that everything a human would actually try is flat. Source: synthetics_findings.md in our research record; ~2.62M ticks/symbol, 2026-05-09 → 2026-06-08; measured spreads Boom ≈ 0.07, Crash ≈ 0.04; spike = single-tick move > 10× median tick (20× threshold gives near-identical counts).