Track 11 · Noise · lesson 3
Why the past looks obvious
11 min
Track 11 · Noise · lesson 3
11 min
Pick any well-known market event of the last few decades. Read what was published in the eighteen months before it.
You will find people who called it, and they will be quoted extensively. You will also find, if you look at the same publications rather than the curated list, that they were surrounded by people calling the opposite, and that the two groups were about the same size and roughly as confident as each other.
The event looks inevitable now. At the time it was a disagreement.
Once you know how something turned out, you cannot reconstruct not knowing. The memory of your earlier uncertainty is rewritten by the answer, and the rewrite is not available to introspection — it feels like recall.
The effect is measurable: people asked to predict something, then asked afterwards what they had predicted, systematically report an earlier belief closer to what happened than the one they actually wrote down. They are not lying. The written record and the memory disagree, and only one of them was made before the answer existed.
The past looks obvious for two separate reasons, and they need different corrections.
Hindsight is the rewriting of your own earlier state. You knew it, roughly, in retrospect. This one is about memory.
Selection is what happens to the record. The people who were right get quoted, reprinted and invited back; the people who were wrong are not re-published, and often quietly stop mentioning it. So the surviving commentary about any past period is systematically more accurate than the commentary actually available at the time.
Put together, these two produce a past that appears far more legible than the present, which then makes the present feel uniquely uncertain and confusing. It is not. It is the only period you are seeing unedited.
Predict
This is not merely an interesting quirk of memory. It has three consequences that cost money.
It makes past decisions look worse than they were. A choice that was reasonable under genuine uncertainty gets reviewed by a person who now knows the answer, and looks negligent. That review teaches the wrong lesson, and teaches it convincingly.
It makes the future feel more knowable than it is. If the past was legible, the present ought to be too, so the failure to see clearly now reads as a personal deficiency rather than as the normal condition. That gap is exactly where someone selling certainty finds their customers.
It destroys the record you would need to calibrate. Since your memory of what you believed is edited to match what happened, you cannot assess your own forecasting by remembering. The only defence is to have written it down, and almost nobody has.
Not the same thing
The information needed to see it was available, and a method would have found it.
A coherent story can be told afterwards about why it had to happen.
Which is which? Put each one on a side.
A rule written in 2010 and tested only on data from 2015 onwards
An explanation of last year's move, published this year, that fits it perfectly
A strategy tuned until it would have performed well over the last twenty years
There is a deeper version of this, and it is worth stating even though it is uncomfortable.
Human beings understand sequences of events through stories, and a story requires causes. So any account of what happened will contain causes, whether or not causes were present. A period in which prices moved for no reconstructable reason cannot be told as a story, so it does not get told — it gets a reason attached and then it can be.
This means the supply of explanations is not limited by the supply of explicable events. It is limited by the demand for explanations, which is enormous and constant. Every day something moves and every day somebody explains why, at the same rate regardless of whether the day contained anything explicable.
Check
Anything can be tuned to fit a known series. Enough parameters and enough attempts will produce a rule that would have performed beautifully over any particular stretch of history, and the tuning is often done in good faith by someone who genuinely believes they are discovering a pattern.
The test that separates a discovery from a fit is whether the rule was fixed before the data it is judged on existed. That is a hard test to pass and an easy one to accidentally fail, because a rule adjusted twice after seeing the results has been fitted even if no single adjustment felt like fitting.
Track 12 returns to this from the other side: it is what makes a performance record so easy to construct and so difficult to interpret.