Stochastic Momentum Index: How to Read the SMI

The Stochastic Momentum Index is a momentum line that swings around zero rather than climbing a 0-to-100 scale, and it arrives with a second line — a smoothed copy of itself — whose crossings are the event most traders actually act on. Vike computes it as batch_smi_ergodic, one of the momentum family in our engine.

Two entirely different indicators answer to those three letters. Sorting that out comes later on this page; first, the arithmetic, then the crossover rule, then 120 real signals scored against real forward bars — the ones that lost included.

What the Stochastic Momentum Index measures

Strip the smoothing away and a single ratio is left underneath: how much of the recent movement went one way.

Net change sits in the numerator, so a day up and a day down cancel each other out. Absolute change sits in the denominator, where nothing can cancel, because every bar contributes its full size no matter which way it closed. Divide the first by the second and the result reaches +100 only if every close in the window rose, and −100 only if every one fell. Chop, with equal force in both directions, sits at zero.

That is the entire idea. Momentum here is not speed — it is agreement.

The exact SMI formula our engine runs

Most write-ups stop at "double-smoothed momentum" and move on. Here is what actually runs.

Take each bar's close-to-close change and keep two copies of it: the signed value and its magnitude. Smooth both with a 20-period exponential average. Then smooth each result a second time, over 5 periods. Divide the twice-smoothed signed series by the twice-smoothed magnitude series, multiply by 100, and you have the SMI. Its signal line is a further 5-period average of that output.

How Vike computes it. batch_smi_ergodic builds delta and abs_delta from consecutive closes, runs a 20-bar EMA and then a 5-bar EMA across each, and returns 100 × ema2(delta) ÷ ema2(abs_delta) wherever the denominator is non-zero. Each pass only consumes values the pass before it defined, so the warm-ups stack: the SMI line starts on bar 23 and its signal on bar 27. Everything earlier comes back as no value at all, never as a partial number dressed up as a real one.

I ported that rule out of the crate source and then checked it against the Python original the crate was itself ported from. Both agree to the last bit across fourteen symbols and a thousand daily bars apiece, which is the only reason the figures below are worth reading.

Reading the SMI against its signal line

Zero is the axis that carries the meaning. Above it, closes have been net positive across the smoothed window; below it, net negative.

Extremes have to be calibrated rather than assumed. Across the sample I measured — fourteen liquid pairs, a thousand daily bars each — the fifth percentile sat at −38.8 and the ninety-fifth at +42.7, so the familiar ±40 lines really do fence off the outer tenth of readings. Observed values ran from −76.9 up to +88.5. Nothing came close to the theoretical bound, because twenty consecutive one-way closes do not happen in a liquid market.

Its signal line supplies the other half of the instrument. A cross above it means the twice-smoothed momentum has turned faster than its own average, which is the earliest arithmetic evidence that the balance is tipping. Crossings around zero are noise, and I skip them without looking. Crossings deep inside a band are the ones that earn a second look.

How to trade the SMI crossover

Here is the rule I tested, stated tightly enough that it can be proven wrong: buy when the SMI crosses above its signal while below −40, sell when it crosses beneath its signal while above +40.

Across 13,454 bars where both lines existed, that fired exactly 120 times — roughly one signal per 112 bars, which rules it out as the backbone of any daily routine. Fifty-two were bullish, sixty-eight bearish. Scoring each one first-touch over the following ten bars, a 6% target against an 8% stop, and awarding the loss whenever a single bar tagged both levels, the long side came out 34 wins to 12 losses and the short side 39 to 25.

Both look respectable until the payoff is priced in. Risking eight to make six needs 57.1% simply to break even before fees, which leaves the short side's 60.9% inside the noise and the long side's 73.9% with genuine room above it. A hit rate quoted without its payoff is decoration.

Confirmation from price still helps. A hammer printing on the same bar as a bullish cross is two independent claims about one moment, and I use that pairing far more than either signal alone.

Where the SMI fails

Now the uncomfortable part, and both failures shown here are real signals this detector produced.

The worst failure mode here is not an incorrect signal. It is a correlated one. Around 2026-02-07 the rule went long on thirteen of the fourteen symbols inside a three-day window; ten hit the stop and not one reached its target. Whoever held "thirteen independent setups" that week was holding a single trade in thirteen sizes.

Double smoothing causes the second problem, and it is structural rather than fixable. Buying a clean line costs you arrival time, so by the point the SMI has bent far enough to cut its own average, a sharp reversal is frequently several bars old already. The mistake I kept repeating was treating a late cross as a fresh one. Now it confirms a move I already had reason to expect, or it gets ignored.

SMI Ergodic is not the 1993 SMI

Platforms muddle these two constantly, and the confusion is expensive.

William Blau published both of them. His 1993 Stochastic Momentum Index measures the close against the midpoint of the recent high-low range, which makes it a cousin of the ordinary stochastic oscillator. His later ergodic form — the one implemented here — divides net change by absolute change and never inspects the range at all. One author, one abbreviation, two different inputs.

They track each other closely enough to be mistaken and diverge often enough to matter. Check which one you have by its parameters: a %K length and a %D length mean the 1993 version, while a long and a short smoothing period mean this one.

Frequently asked questions

What are the best SMI settings? Twenty, five and five are the defaults, and every figure on this page was measured on them. Shortening the long period makes the line twitchy and multiplies crossings, and each extra crossing is another chance to be wrong.

Is the SMI better than RSI? Neither is better; they answer different questions. RSI is bounded between 0 and 100 and needs a threshold you pick, whereas this line centers on zero and carries its own reference, so it emits events rather than states.

Why do the first bars have no value? Two stacked exponential averages need 23 bars before any SMI value exists, and 27 before its signal does. Returning nothing there is deliberate.

Does the SMI ever reach +100? That requires a window in which every single close rose. Across 13,678 real readings the highest observed was 88.5, so treat ±100 as geometry rather than something you will see.

Which indicator pairs well with it? Something that measures a different thing — the Relative Vigor Index reads each bar's open-to-close position instead of its close-to-close change, so the two disagree in useful ways.

This is educational material, not financial advice. Every number here is measured on past bars, past behavior generalizes poorly, and trading carries real risk of loss — size any position so that being wrong stays affordable.

Read the Relative Vigor Index breakdown