It's Just Noise

Call the trend — then watch it betray you.

No real system runs on clean numbers. Every loop is jostled by chance — weather, moods, a customer who happened to call, a coin that happened to land heads. And here is the thing our pattern-hungry brain cannot accept: random shocks, fed through an ordinary balancing loop, produce runs. Three good weeks in a row. A line that drifts up for a while and then down. Slopes, streaks, “momentum.” They look exactly like signal — like a story is unfolding — and underneath there is nothing at all but a flat goal and a fair coin.

So don't just look at it — bet on it. Below is a stock held at a perfectly flat target, plus noise, and no real trend whatsoever. You see it up to “now” and can read the run as plainly as anyone. So commit: will it end the stretch higher or lower than it stands now? Then reveal. Do it eight or ten times and watch your hit-rate — it settles at a coin-flip, because the trend you were so sure of was never there.

This is why a run of good quarters gets a CEO a book deal and a run of bad ones gets them fired, when both may be weather. It is why we chase the hot fund, fire the cold manager, and change a diet after one heavy weigh-in. The discipline is brutal and simple: before you explain a move, ask whether a flat line plus noise would have produced it anyway. Most of the time, it would.

“Don't be surprised when a system exhibits a wide range of behaviors. Variability is normal; the search for the cause of every fluctuation is often a fool's errand.”

after Donella Meadows · on systems & variability

If reading stories into noise is harmless curiosity, the next step is where it gets expensive: acting on each wiggle. Adjust the system every time chance nudges it, and you don't cancel the noise — you amplify it. That's Game Two.

React to every random miss and you double your own chaos.

You have a process that sits on its target, jostled by noise. A reading comes in a little high. The obvious move — the responsible-feeling move — is to nudge the setting down to cancel it. Next reading's a little low, so you nudge back up. You are working hard, staying on top of it, correcting every miss. And you are making it worse. Because the miss was random, your correction isn't cancelling a real drift — it's a fresh disturbance laid on top of the next random one. W. Edwards Deming proved this with a funnel and a marble and a target on the floor: the team that adjusted after every drop scattered twice as wide as the team that bolted the funnel down and never touched it.

Same noise, same target, two policies on the same screen. The faint line is hands-off — the loop left alone to absorb its own shocks. The bold line is tampering — adjusting to cancel each last miss. Turn up how hard you tamper and watch the bold line fan out past the faint one. The variance number is the verdict: tampering doesn't shrink it, it grows it.

Tampering is everywhere a well-meaning hand meets a noisy number: re-tuning a machine after each part, rewriting strategy after each quarter, adjusting a child's plan after each test, trading on each day's move. The skill that looks like laziness is the hard one — to know the difference between a signal worth acting on and a wobble worth ignoring, and to keep your hands off the second kind. Don't steer the noise. Only steer the trend.

“If the factory is variable, don't tamper with it — improve the system that produces the variation. Adjusting to noise is not control; it is the addition of more noise.”

after W. Edwards Deming · the funnel experiment

Tampering hurts because, in a balancing loop, the noise was going to wash out on its own. But put that very same noise inside a reinforcing loop and it never washes out. It compounds — and decides the winner. That's Game Three.

A balancing loop forgets luck. A reinforcing loop keeps it forever.

We are taught that randomness averages out — that over enough trials, the good and bad luck cancel and the truth shows through. That is true, but only inside a balancing loop, where every shock is pulled back toward the goal and yesterday's luck is erased by today's correction. Run many noisy balancing loops and they all stay huddled in a band around their target; you cannot tell the lucky from the unlucky, because the loop won't let early luck matter. Now run the same noise through a reinforcing loop — where each step multiplies the last — and luck stops cancelling and starts compounding. An early good shock makes the base bigger, so the next shock acts on more, and the paths fan apart into winners and losers that will never reconverge.

Flip between the two loops below. Each draws two dozen paths from identical random shocks. The balancing loops stay in a tight band — order restored, luck forgotten. The reinforcing loops explode into a fan — a handful run away to the top on nothing but a lucky start, the rest are left behind, and no amount of later skill closes the gap. Same randomness; two completely different fates, decided by the wiring.

This is the quiet link between noise and everything else in this family: in a mean-reverting world, ignore the luck and wait. In a compounding world, the luck is the outcome — so the move is to take many small uncorrelated draws rather than one big one, survive the bad shocks long enough to stay in the game, and never mistake a lucky path for a skilful one. The loop decides whether your randomness is forgotten or made permanent. Read the loop before you read the result.

“Remember, always, that everything you know, and everything everyone knows, is only a model. Get your model out there where it can be viewed. Invite others to challenge your assumptions and add their own.”

Donella Meadows · Thinking in Systems