Part XI · Swing trading · Chapter 11B

Conditional mean reversion

The central claim, and the conditions it needs

1 · The thesis, stated with its conditions

The dashboard of Chapter 11A tells you what regime you are in. This chapter is about what to do when it says the regime is mean-reverting. That means three things hold at once: the macro layer allows mean reversion as the prior, structure puts the market at a pullback level, and dealer gamma positioning makes damping intraday flow likely. The conditional thesis is not "SPY mean reverts" but "SPY mean reverts when measurable conditions hold". The dashboard's four checks are exactly those measurable conditions.

As a universal claim about SPY, mean reversion is empirically false. Chapter 3B showed why a pooled fit can hide it: the sign of the effect changes with the regime. The classic statistic compares the average next-bar return after an up stretch with the average after a down stretch. Write $m_{\text{up}}$ for the mean return of the bar after a trailing $N$-bar return above zero, and $m_{\text{down}}$ for the mean after one below zero. The statistic is their difference:

A positive $D$ is momentum: up stretches are followed by higher returns than down stretches. A negative $D$ is mean reversion: down stretches are followed by the higher returns. On SPY the sign of $D$ depends on the regime. Sort the days by VIX level, by the sign of GEX or by the direction of credit spreads, and the sign flips. A trader who claims to be "a mean-reversion guy" is making a regime assumption without realizing it.

The figure works the example with its numbers free. Lower the calm share of days, and the pooled bar falls further below the cost line while the calm bar stays above it. Raise the cost past 0.10%, and neither rule pays.

calm share of days

80%

round-trip cost

0.05%

calm effect

+0.10%

stressed effect

−0.15%

stress: 20% of days, 30% of down weeks

pooled after cost = −0.025%

calm only after cost = +0.05%

Conditioning pays. Taken after every down week, the rule does not clear its cost. Taken only in the calm regime, it clears it by 0.05% a trade.

2 · Writing the conditions down

The conditional framing avoids the trap. You write down the conditions under which you expect the edge to be positive, and you check them. If any condition fails, you do not take the trade. Some condition fails on most days, which is the system working, not a sign that it is broken.

3 · Three nested horizons

The framework is a hierarchy in which each layer answers a different question, runs on a different clock and uses different inputs. A signal at one layer never overrides a contradicting signal at a higher layer.

Layer 1, the macro regime, sets the prior. Its inputs are the equity drift, read from the slope of the 200-day SMA, the VIX level and term structure, and credit and rates stress: HY OAS, IG OAS and the MOVE index. It also reads the yield curve, as 10Y−2Y and 10Y−3M, the dollar and FX (DXY, USD/JPY) and the composite signal score of Chapter 11G. In its default states, positive drift with contango and tight credit means mean reversion is on. An inverted curve with widening credit and backwardation means mean reversion is suspended.

Layer 2, market structure, finds the level and the pattern. Its inputs are the Weinstein stage (Chapter 11E), the location against key moving averages (the 10-, 21-, 50- and 200-day EMAs), pattern detection (Chapter 11D) and relative strength against a benchmark. It decides whether this specific setup sits where the macro edge can be expressed cleanly.

Relative strength is the price of the stock or index divided by the price of a benchmark, followed over time. A stock that gains 8% while SPX gains 3% sees its relative strength rise by $1.08/1.03 - 1 = 4.9\%$. "Relative strength positive" means the ratio is rising. Chapter 11D gives the Mansfield version, which compares the ratio with its own average. Relative strength has nothing to do with the relative strength index of Chapter 11C, despite the name.

Layer 3, dealer gamma and path, sets the execution context. Its inputs are net dealer GEX (Chapter 8A), the distance to the gamma flip, how the VIX behaves intraday, options skew and the time of day. It decides whether today specifically supports the trade, or whether mechanical flow works against the position.

4 · The decision tree

The layers combine into a decision tree that you run in order:

1. Macro regime supportive?
   (high composite signal, contango,
    positive drift)
   If NO:  no trade.
   If YES: continue.

2. Structure says BUY?
   (Stage 2 pullback to a defined level,
    pivot intact, relative strength positive)
   If NO:  no trade.
   If YES: continue.

3. Gamma regime says the PATH is acceptable?
   (positive GEX above the flip, OR a clear
    setup for an amplification trade below it)
   If NO:  skip.
   If YES: take the trade.

4. Size for the loss budget (Chapter 13C),
   pre-commit the invalidation (Chapter 13C),
   record the trade in your journal.

Most days you stop at step 1 or step 2, which is correct for a framework designed to filter aggressively. The Dealer Mechanics Lab adds that trades surviving all three gates should be sized larger than trades that fail any of them. Since the tree takes no trade that fails a gate, read this as: concentrate your size in the few survivors rather than spreading it across every setup you see.

Sizing the survivors larger has a formal reason in the Kelly criterion, whose bet grows with the edge per unit of variance (Chapter 3D). A trade with a higher expected return earns a larger fraction of capital. A fractional Kelly bet keeps that ordering at a safer scale. The gates raise the expected return of the trades they pass, which is why the survivors deserve more.

Where the Edge Book disagrees

The Edge Book is far less confident that this edge is open to a retail account. It files short-term reversal over 1 to 5 days, liquidity provision after sharp moves, among the microstructure edges (Chapter 1G). It calls those essentially institutional-only: the edge per trade is usually 1 to 5 bps, which retail commission and slippage alone exceed. It also treats the sign of GEX, the core input of Layer 3, as low-confidence context worth only a fraction of one filter's weight (Chapter 10H).

The Dealer Mechanics Lab's case rests on the conditioning itself. It points to positive-gamma dealer hedging, which sells rallies and buys falls, as the mechanism behind reversal in calm regimes (Chapters 3B and 8D). It also holds each trade for 2 to 10 days rather than minutes. Neither source reports the conditional edge per trade after retail costs. Only your own out-of-sample test can supply the number that decides the question: if the conditioned edge does not clear your round-trip cost, the Edge Book's objection stands.

5 · Grading the evidence

Not every claim in this framework carries the same weight of evidence. The Dealer Mechanics Lab tags each with a tier:

One Tier A example needs its scope. Ni, Pearson and Poteshman studied optionable single stocks from 1996 to 2002. On each expiration date they measured at least 16.5 basis points of return distortion, on average. They did not study the index. For S&P 500 futures, the evidence is Golez and Jackwerth (2012), who found a pull toward the strike worth at least $115 million of notional per expiration day.

Tier C is real but heuristic, which leaves the rigorous formalization to you. Chapters 11D and 11E do some of it, turning pattern language into explicit thresholds. Always tag in your own notes which tier each claim sits in. The tier does not disqualify a claim, but it sets the prior for how heavily to weight it.

The Edge Book grades evidence on a different axis. Its evidence tiers T1 to T3 grade how strong and durable the evidence for an edge is (Chapter 10A), while Tiers A to C grade the kind of source. The two can diverge: a peer-reviewed paper can document an effect that has since decayed, which Chapter 10I shows happening more than once.

6 · Tagging the horizon

Every signal also gets a horizon tag: a macro regime call works over weeks to months, a structural pattern over 2 to 10 days, and a gamma-driven intraday move resolves in hours. Mixing horizons is the second-most-common failure, after mixing tiers. People see a 5-minute breakout and infer a 6-month thesis, or hold a 6-month thesis and trade it on a 5-minute chart.

The arithmetic of Chapter 1A shows why a thesis at one horizon says little at another. Over a horizon $h$ in years, a drift $\mu$ moves the price by about $\mu h$, while the noise has a standard deviation of $\sigma\sqrt{h}$. Their ratio is the signal-to-noise ratio of the thesis at that horizon:

The discipline is to write the horizon of every trade idea down before you enter. If the horizon you stated does not match the indicators you are using, fix one or the other before you click buy.

7 · What the next chapters build

The rest of the machinery comes in five chapters:

• Chapter 11C builds the indicator foundations: RSI, MACD, ATR, ADX, Bollinger Bands and Donchian Channels. It asks which feature of the data each measures, and which combinations are genuine confluence rather than reparameterization.
• Chapter 11D builds a pattern library of the volatility contraction pattern (VCP), the cup and handle and the bull flag, as detectors with explicit geometric thresholds.
• Chapter 11E gives Weinstein's four-stage classifier and its overlap with GEX regimes.
• Chapter 11F gives Elder's triple screen across timeframes. The higher timeframe (HTF) sets the direction, the intermediate one (ITF) the timing and the lower one (LTF) the trigger, with confluence as a discipline against impulse.
• Chapter 11G aggregates an eight-dimensional regime scorecard into a single bias number. Its dimensions are volatility, credit, rates, the dollar, commodities, global markets, trend and sentiment.

8 · Check yourself