Downswings - How Long Can a Legitimate System Run Bad Before You Should Doubt It?

[SharpEddie47]

The question every serious bettor eventually faces and most don't have a prepared answer for.

2018. My most significant downswing on record. Eleven weeks. NFL season. The model was producing selections. I was executing correctly. I was losing consistently.

By week eight I'd started questioning everything. Were the edges I'd identified real. Had the market changed without me noticing. Was my sample size from previous profitable years actually sufficient.

By week ten I was on the verge of making systematic changes to a methodology that had worked for seven years prior.

Didn't make the changes. Held the system. Weeks twelve and thirteen recovered most of the losses. By the end of the season the P&L was mildly negative. Within variance.

The question I couldn't answer during week eight: was this legitimate variance or real edge disappearance.

The math says a genuine 54% system at evens will produce a run of ten or more consecutive losses roughly every eight hundred bets. The math is cold comfort when you're in it.

How do people think about this. What's the threshold where doubt becomes reasonable rather than just emotional.

 

[Klaus]

Have a pre-committed framework for this specific question.

Before the season: I calculate the expected maximum drawdown for the model given historical edge and variance parameters.

The calculation produces: given a genuine 3.8% edge and the variance distribution of my historical results, I expect to see a maximum drawdown of X% of bankroll during any given season with 90% probability.

If my actual drawdown exceeds that expected maximum: I investigate the model.

If my actual drawdown is within the expected range: I continue executing the system.

The decision rule is pre-committed before the bad run begins.

Making the decision during the bad run is making it in the worst possible psychological conditions.

The pre-committed rule removes the decision from the emotional state.

This season: experienced a seven-week losing period. Within the pre-calculated expected drawdown range. Continued executing. Recovered in weeks eight through twelve.

The framework held because I'd built it before I needed it.

 

[FadeThePublic]

Three significant downswings in fifteen years.

Each one I was convinced I'd lost the edge.

Each one eventually resolved as variance.

But the honest thing is: I had no reliable way to distinguish variance from real edge disappearance while I was inside the bad run.

The retrospective clarity is easy. "That was just variance." The prospective certainty is unavailable.

The specific thing I've found useful: asking what would have to have changed in the market to produce this bad run if it isn't variance.

If the answer requires specific identifiable changes I can test: worth investigating.

If the answer would require the entire market structure to have shifted in undetectable ways: almost certainly variance.

The bad run that has a coherent external explanation is worth taking seriously.

The bad run that would require implausible market changes to be explained as edge loss: probably variance.

 

[TaffyTipster]

Don't have a mathematical framework.

Just white-knuckle through.

Bad runs feel exactly the same whether they're variance or real edge loss.

There's no sensation that distinguishes them.

The only practical thing I do: keep the stakes consistent during bad runs.

The temptation to reduce stakes or increase stakes is always there.

Reducing: feels prudent. Is actually locking in the losses at exactly the wrong time.

Increasing: feels like recovery mode. Is actually compounding the problem if it is a real edge problem.

Flat stakes through bad runs is the only rule I've consistently applied.

Not because I understand the statistics. Because I've seen what happens when I deviate.

 

[CoachTony_Bets]

The coaching parallel is exact.

Every coach goes through a losing stretch. The question is what caused it and what to do about it.

If the opponent quality was higher: variance from schedule. Continue the system.

If the players stopped executing properly: system problem. Adjust.

If the scheme has become predictable: real strategic problem. Change.

The diagnosis requires honest analysis not emotional response.

The worst coaching decision I've seen repeatedly: changing the system during a variance-driven losing run.

New scheme causes confusion. Execution suffers. More losses. Confirms the coach's fear that something was wrong.

The losing run created the change. The change created more losses. Neither was necessary.

I've seen this happen to excellent coaches who made one emotional decision under pressure.

Done the betting equivalent myself.

 

[oli_sussex]

At the exchange professional traders had a specific protocol for downswings.

Mandatory review at pre-defined drawdown thresholds.

Not optional. Not when you felt like it. At the specific number.

The review asked specific questions.

Has the market structure changed in ways visible in the data. Is the loss distribution consistent with the model's expected variance. Are the specific markets that are losing different from the ones that are winning in identifiable ways.

If the review found no structural explanation: continue trading. The drawdown is within expected parameters.

If the review found structural changes: pause. Investigate. Potentially adjust.

The review wasn't about whether you felt confident. It was about what the data said.

Most retail bettors review when they feel bad. The timing is wrong. You should review at pre-defined thresholds regardless of how you feel.

 

[ParlayPrincess_88]

I've had bad runs and immediately changed what I was doing.

Three weeks of parlay losses. Changed my selection process. Started picking differently.

The new selections also lost for two weeks. Changed again.

Probably was all variance from the beginning and I was changing a process that didn't need changing.

Now I've got no consistent baseline at all because I've changed too many times.

The variance never got a chance to resolve because I kept interrupting it with changes.

 

[SharpEddie47]

Princess describing the exact failure mode that makes variance so dangerous.

The system changes during a bad run create a new problem.

Even if the original system was fine, you'll never know now.

You've reset the sample. The new system needs its own validation. The bad run's cause becomes permanently ambiguous.

The cost of changing too early isn't just the losses you'd have recovered.

It's the analytical certainty about whether the system worked that you'll never recover.

 

[DublinDegen]

never had a system to doubt during bad runs...

just bad runs...

the question for me was never "is the system broken" it was "when does this stop"...

the answer was always "keep going and it will stop eventually"...

which is the answer that makes sense for someone with a real edge and terrible advice for someone without one...

both the person with a legitimate edge and the person with no edge experience bad runs...

the legitimate edge person should hold through it...

the no-edge person should stop...

but the experience of the bad run is identical...

there's nothing in the feeling of losing that tells you which category you're in...

that's the specific thing that's kept me going through bad runs that should have stopped me...

 

[FadeThePublic]

Conor identifying the most important thing in this thread.

The phenomenology of the downswing is identical for the legitimate edge bettor and the no-edge bettor.

The losses feel the same.

The uncertainty feels the same.

The temptation to change or continue feels the same.

The decision to hold or fold requires knowing which category you're in.

Knowing which category you're in requires the kind of documented evidence that most people don't have before they need it.

 

[Klaus]

The pre-committed framework addresses this problem partially.

If you've done the work before the bad run: you know the expected variance range. You know what would require investigation. You have a decision rule that doesn't depend on feeling.

If you haven't done the work before the bad run: you're making real-time judgments with impaired cognition in the worst possible conditions.

The person who hasn't built the framework before the bad run can't build it during the bad run.

They're making decisions without the necessary information.

Conor's situation: no framework, no baseline, no way to distinguish variance from absence of edge.

The correct response to that situation is to stop until you have a framework.

The actual response to that situation is rarely stopping.

 

[ThePuntingProf]

Three distinct experiences of downswings over thirty years.

The first: 1998-1999. Extended bad run in English football markets. Convinced I'd lost whatever analytical advantage I'd developed. Made significant methodology changes.

Eventually determined through retrospective analysis that the original approach was probably fine and the changes were unnecessary. The bad run was within variance for the sample I'd accumulated.

The second: 2009. Bad run that corresponded with genuine market changes. The operators in the markets I was using improved their pricing significantly in that period. The bad run had a real cause. The methodology required genuine adjustment.

The third: 2020-2021. COVID disruption. The patterns I'd identified over thirty years didn't apply to matches played behind closed doors. Real cause. Temporary. Adjusted and returned to previous approach when conditions normalized.

Three bad runs. Two were variance. One was real.

I couldn't distinguish them while inside any of the three.

The retrospective analysis was clear. The prospective clarity was unavailable.

I had better luck with the determination than better judgment.

 

[SharpEddie47]

Prof's three-case dataset is the most honest description of the problem.

Two out of three were variance. Would have been fine to hold.

One was real. Required change.

Couldn't distinguish during the event.

The decision to hold or change was made with incomplete information in all three cases.

Got one of the three meaningfully wrong.

The one wrong decision cost less than systematic overcorrection would have.

Holding when uncertain, changing only when the evidence was overwhelming: probably the right heuristic even without a formal framework.

 

[FadeThePublic]

The holding heuristic works better than the changing heuristic over time for a specific statistical reason.

False positives from changing during variance: you interrupt a legitimate edge unnecessarily. Cost is validation loss and potential performance loss from an inferior new approach.

False positives from holding through real edge loss: you continue with a broken approach. Cost is continued losses until evidence accumulates.

The second error is more recoverable than the first.

A broken approach reveals itself eventually with continued losses.

A changed approach might appear to work for a while even if the original was better.

Erring toward holding is statistically better for most distributions of outcomes.

Not comfortable. Statistically better.

 

[CoachTony_Bets]

The most important practical advice in this thread hasn't been said explicitly.

Build the variance parameters before you need them.

Simulate your expected drawdown distribution before the season starts.

Decide at what threshold you'll investigate versus hold.

Write it down. Commit to it publicly if possible.

Then when the bad run comes: apply the pre-committed rule.

The decision made in advance is better than the decision made during the bad run in every way except that it requires doing work before you think you need it.

Nobody does the work before they need it.

Then they need it and they don't have it.

 

[Klaus]

Tony's observation is accurate.

The preparation required to navigate downswings correctly must be done before the downswing.

During the downswing the cognitive resources required for that preparation are consumed by the psychological experience of losing.

The person who needs the framework most is the person least able to build it.

The solution is building it at the beginning of each betting year as standard practice regardless of current results.

Most bettors don't do this.

Most bettors navigate their most important decisions with their worst available cognition at their most emotionally compromised moments.

 

[DublinDegen]

that last part...

navigating the most important decisions with the worst available cognition at the most emotionally compromised moments...

that's not just about downswings...

that's the whole thing...

the bet that costs the most is always the one placed when you're least equipped to place it...

the downswing is just one version of that...