Noise vs. Signal: Why Checking More Often Makes You Wrong More Often

Here's a counterintuitive result from Fooled by Randomness that I think about constantly.

The more frequently you monitor a result — a portfolio, a business metric, a relationship, your health — the worse your signal-to-noise ratio gets. At shorter time intervals, what you observe is increasingly dominated by random fluctuation with no predictive content. At longer time intervals, the underlying signal becomes visible.

Checking more doesn't make you better informed. It makes you better informed about noise.

The Dentist's Portfolio

Taleb's example is a dentist with a balanced, well-constructed portfolio — expected annual return of 15%, annual volatility of 10%. These are Taleb's illustration numbers, but the ratio matters more than the specifics. Good expected outcome, meaningful short-term variance.

Now watch what happens to signal-to-noise as we shorten the observation interval:

At annual resolution, the dentist has roughly a 93% chance of seeing a positive year. What he observes is mostly signal — the underlying return is doing most of the work.

At monthly resolution, the probability of a positive month drops to about 67%. One in three months is negative, driven entirely by random variance around a positive expected return.

At daily resolution, the positive probability drops further, to around 54%. Slightly better than a coin flip.

At minute-by-minute resolution, the probability of a positive minute is essentially 50.02% — functionally a coin flip, at which point the observation contains no useful information about the underlying portfolio at all.

The dentist's portfolio is unchanged. His volatility is unchanged. His expected return is unchanged. What changed is the resolution at which he's observing it — and at sub-daily resolution, almost everything he sees is noise.

The Asymmetry Makes It Painful

It would be tolerable if seeing noise were neutral. But noise is not emotionally neutral. Losses hurt more than equivalent gains feel good — the classic loss aversion finding from Kahneman and Tversky. So the dentist checking his portfolio daily doesn't just see noise; he experiences the noise asymmetrically. Each negative day stings more than each positive day satisfies.

Over a year of daily checks: - Roughly 54% of days are positive (minimal satisfaction) - Roughly 46% are negative (significant pain) - Net emotional experience: negative, even for a portfolio with a strongly positive underlying return

The high-frequency monitoring doesn't just add noise to the dentist's information. It produces a biased emotional experience that diverges from the actual trajectory of the investment. He's suffering on a winning position because he's observing it at a resolution where losing is the frequent experience.

The obvious behavioral consequence: this triggers emotional decisions that interrupt a compounding process that was working fine. The dentist who checks daily is much more likely to sell during a drawdown, to "de-risk" after a string of negative days, or to make changes based on short-period performance that has no predictive content for the long-period outcome.

Where Else This Applies

The insight extends well beyond investing.

Business metrics. Checking daily active users, revenue, or engagement rates daily exposes you primarily to noise. Day-to-day variation is dominated by weekday effects, weather, news cycle, minor algorithm changes. Weekly is better. Monthly better still. The underlying signal — whether the product is growing or declining — only becomes clear at longer intervals.

Relationship quality. A couple that rates their relationship weekly will observe the noise of individual weeks: the bad Tuesday, the good weekend, the stressful project at work. A couple that asks "are we moving in the right direction over six months?" is observing at a resolution where the signal is more legible.

Health tracking. Daily biometric monitoring (resting heart rate, sleep score, readiness) introduces noise at a level where random biological variation dominates. Day-to-day sleep scores vary substantially from temperature, caffeine timing, and minor lifestyle variation. The person treating a 62 readiness score as information is usually responding to noise. Annual bloodwork and functional metrics are the resolution where signal predominates.

The Right Inspection Interval

The rule that follows: match your inspection interval to the underlying signal's characteristic time scale.

For decisions with long time horizons (long-term investments, business strategy, personal development), weekly or daily monitoring is almost certainly observing primarily noise. The signal takes months or years to emerge. Monitoring at sub-signal resolution creates the illusion of information while producing emotional responses to meaninglessness.

For decisions with short time horizons (a live-fire trading position, a customer experience metric where real-time intervention is possible), shorter monitoring intervals may be appropriate. But even here, the question to ask is: at this interval, is what I'm seeing actionable signal, or random variance I'll respond to incorrectly?

The financial media is the institutional case of wrong-resolution monitoring. News outlets report markets at minute and hour resolution. The content this produces is overwhelmingly noise — daily explanations for daily fluctuations that are random. Narrative is imposed on coin flips. The result: investors who consume financial media heavily are systematically informed about randomness and systematically misled about signal.

Taleb's discipline: not reacting to low-frequency fluctuations. Extending the inspection interval until the underlying direction is visible. Tolerating the short-term variance that comes with not observing every tick, in exchange for a cleaner view of the long-term trajectory.

For the full framework, read Fooled by Randomness: How Luck Masquerades as Skill.