Picking Up Pennies Ch. 5: Sizing for the Regime You Don't Know You're In

Investment Background

risk-models-portfolio-construction ch01–02, already in this library, builds the Kelly Criterion carefully: given a known, correctly-estimated edge, there is a mathematically optimal fraction of capital to risk, and overbetting even a real edge causes ruin. That chapter's entire framework assumes the edge is known. Chapter 4 of this book just established that, for any systematic or volatility-selling strategy, you cannot be certain the edge is still there at all. This chapter sits at the intersection: how do you size a position when the input the Kelly formula needs — a reliable edge estimate — is exactly the thing under doubt?

The Wall Street Translation

Why Standard Sizing Logic Breaks Down Here

The Kelly Criterion, and every sizing framework built on it, requires an edge and a confidence level as inputs. For a strategy that may have already crossed into a dead regime — as Chapter 4 describes — the honest confidence level in the edge is not a fixed number you can plug in. It degrades continuously and invisibly the longer a regime persists undetected, and you have no reliable signal for how far that degradation has already gone.

This means position sizing for negative-convexity, short-volatility-shaped strategies needs an additional discipline layered on top of Kelly, not a replacement for it: size for the case that your edge estimate is already stale, not only for the case that it is current.

The Math of Sizing Under Regime Uncertainty

Here is the mechanism, with numbers.

Suppose a strategy's backtested edge would justify, under a standard fractional-Kelly approach, risking four percent of capital per position. Standard practice — already covered by risk-models-portfolio-construction ch02 — is to run at a fraction of full Kelly, say half, to control for estimation error: two percent of capital.

This chapter adds a second discount, specific to strategies whose payoff shape is short-volatility or short-tail: because Chapter 1 showed these strategies produce long, confidence-building winning streaks specifically at the moment their true edge is weakest, the position size should be further reduced as a function of how long the current winning streak has run — not increased, which is the trader's natural instinct.

Concretely: if the standard fractional-Kelly size is two percent, and the strategy has just completed its eighteenth consecutive winning month, this chapter's rule says treat the confidence interval on the edge estimate as wider, not narrower, than it was at month one — and size down toward one percent or less, precisely when every other signal (the smooth equity curve, the calm client, the temptation to increase allocation) argues for sizing up.

The Rule Runs Backward From Instinct, On Purpose

This is the chapter's central, uncomfortable point: for this specific shape of strategy, a long winning streak should reduce conviction in the size, not increase it. For an ordinary strategy with symmetric, well-behaved returns, a long track record is legitimate evidence of a durable edge and can justify increased sizing. For a strategy whose payoff shape manufactures confirming outcomes regardless of whether the edge survives, the same long track record is close to uninformative, and the natural human response to it — size up — is systematically miscalibrated.

Way of the Turtle, already in this library, describes a related mechanical danger: one lucky oversized bet teaching a trader to keep sizing too large. This chapter's version is subtler: it is not one lucky bet but an entire, apparently disciplined, apparently well-managed winning streak that teaches the wrong lesson, because the streak itself is the strategy's normal appearance right up until the loss.

Executable Trading Rules

1. Run any short-volatility or negative-convexity position at a fraction of the size the Kelly framework would justify from a fresh, current edge estimate — and reduce that fraction further as the current winning streak lengthens, rather than holding size constant or increasing it.

2. Set the maximum position size using the strategy's worst historical loss, scaled up by a safety multiple, not its typical monthly result. Chapter 1's math showed typical results and tail results tell almost unrelated stories for this payoff shape.

3. Write a specific rule for reducing size after a fixed number of consecutive winning periods — decided in advance — precisely because the instinct in the moment will argue the opposite. The rule exists to override the instinct, not to confirm it.

4. Never use an unusually long winning streak as justification for increasing allocation to a short-volatility or systematic strategy. Treat streak length as a call to re-examine the regime assumption from Chapter 4, not as license to add capital.

Relevance to a Retirement Portfolio

No part of this chapter recommends that a retirement investor run a short-volatility strategy, sized carefully or otherwise — the correct sizing for most retirement accounts is zero. Its value is in recognizing the same backward logic when it shows up in a manager pitch or a fund's marketing: a strategy or fund that emphasizes a long, smooth track record as its main selling point is, by this chapter's logic, providing weaker evidence of durability than it appears to be providing, not stronger.

The core, diversified portfolio sidesteps this entire problem by design. It does not require estimating whether a specific edge has survived a regime change, because it is not sized against any single strategy's edge — it is sized as a stable allocation across broad asset classes, rebalanced on a fixed schedule regardless of any one holding's recent streak.

Chapter 6 closes the book with the honest limit of everything covered so far: naming these failure modes does not, by itself, protect anyone from them.