LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.

A market maker does not need the next outcome to land on a single number. They publish a bid, the price they will buy at, and an ask, the price they will sell at, around their best forecast. In a one-mile timing example, the forecast was 6:21, the bid was 6:11, and the ask was 6:31. The run finished at 6:25. Anyone who bet the runner would be slower than 6:31 lost, and anyone who bet faster than 6:11 lost. The quote sat in the middle and kept the spread.

Why it matters

This is the smallest complete picture of a market. There is a forecast, a width that expresses doubt, and a realized outcome. Transaction costs and spreads are why a small forecasting advantage can still lose money. You pay the width before your own edge has a chance to matter. If that bothers you, the alternative in the lecture is not a secret indicator. It is to be the person quoting, which means holding inventory and being wrong in public when the outcome jumps outside the band.

The same arithmetic is why a casino, or a bank that earns a spread, can do well without predicting each customer. The edge is the width plus the odds, not clairvoyance.

The idea in plain words

The forecast has to come from somewhere. A year of training logs contained more than 437 workouts on a Coros watch, exported to Python. Inside that history were 160 runs and 536 miles. The fastest mile in the set was 5:36. One effort was so poor the runner could barely move. Mixing every distance, from a single mile through half marathons, is a weak predictor of today’s all-out mile.

So the average was not “every mile split in every workout.” It used purposeful one-mile attempts, and it weighted recent races more heavily because the runner had been getting faster. That conditional, recency-weighted estimate was 6:21. Confidence was high enough to quote, and not high enough to quote tight. Heart rate, training load, and aerobic fitness were not in the model. Without a richer conditional expectation, the ten-second pad on each side was the honesty tax. A more confident model would have narrowed 6:11 and 6:31.

The race itself is just the random outcome. The clock stopped at 6:25, a best dedicated mile for that athlete in the telling, and still inside the spread. Betting the over, buying at 6:31, lost. Betting the under, selling at 6:11, lost. No other market maker was offering a better price. That is what a one-sided market feels like: you trade at the only quote, or you do not trade.

How to use it

When you see a bid and an ask, locate the mid, the average of the two, and ask whether you have a reason to trade that is larger than half the spread. If your forecast and the mid disagree by less than you will pay to cross, you do not have a trade. You have a preference.

If you are the one quoting, write down what would tighten the spread. More relevant history, a stable relationship, and a model that uses the variables you already believe matter. Then write down what would widen it: a regime change, a thin sample, or a variable you refused to measure. The mile quote was wide because the model was a weighted average of similar attempts, not a physiological model.

What usually goes wrong

Averaging unlike events. A half-marathon pace is not an all-out mile, just as a calm stock and a crash print are not one distribution. The lecture’s first improvement was to condition on the right race, then to up-weight the recent past.

Quoting a false precision. A point forecast with no width pretends the model is sharper than the data. The 20-second window was the admission that nonlinear models and extra covariates were missing.

Forgetting who gets paid when the outcome lands inside the window. Both directional bettors can be wrong at once. That is not a paradox. It is the definition of crossing a spread around a decent forecast.

Questions people ask

Why was the fair value 6:21 rather than the best mile?

The best mile, 5:36, was one tail of 160 runs. The quote used recent purposeful miles, weighted toward the present, because the athlete had changed.

What would have narrowed 6:11 to 6:31?

A model that also used heart rate, recent training load, and fitness. Those inputs were available in principle and left out, so the spread stayed wide.

Who made money when the mile was 6:25?

The quote. Both sides that traded against the spread lost. The result landed between the bid and the ask.