Intraday Mean Reversion Options: Why Signal Quality Drops When You Chase Density

Repository reference: cutebacktests

Abstract

Intraday mean reversion options strategies often fail in a predictable way. The high-quality version barely trades, so the researcher loosens the setup to gain sample size. Trade count rises, but the original edge softens because the additional trades are exactly the ones the selective version was trying to avoid.

This repository's c36 branch makes that failure mode unusually visible. As summarized in Toward The One Piece Of Sharpe, the higher-quality VWAP mean-reversion branch made +16004 PnL on 15 trades with DSR 0.6400 and failed only trades_per_week_ok, while the opportunity-biased version reached 85 trades and +2987 PnL but lost enough quality that it did not replace the higher-quality anchor. That is the core density tradeoff in one compact comparison.

For the signal-density angle, compare this note with VWAP Mean Reversion Signal Quality and Density, Backtesting Engine Loop, and Backtesting Execution Realism. The terms that matter are VWAP deviation, z-score, setup density, entry cutoff, same-bar fill, option spread, quote age, slippage, and rejection rate.

Question

The practical question is not whether a mean-reversion strategy should be selective. It is how much selectivity can be relaxed before the branch stops being the same edge.

That is an especially important question in options work because the monetization layer is already fragile. Entry spreads, time decay, and short holding periods all make mediocre setups more expensive. If the signal quality decays while density rises, the option layer will usually amplify the damage.

Method: Why Intraday Mean Reversion Options Become a Density Problem

The c36 branch is helpful because it expresses the same signal family through a controlled variation in thresholds. The high-quality version uses stronger VWAP-residual excursion requirements, bounded VWAP slope, sigma constraints, stronger relative volume, and shorter time budgets. The opportunity version relaxes those constraints to create more trades.

Both branches are then expressed through quote-aware single-leg options in the 0-2DTE window. That means the comparison is not being clouded by a completely different instrument structure. The main variable is the strictness of the entry definition. This is important because it turns a familiar qualitative question into a measurable one: how much of the original mean-reversion edge survives when you widen the sample?

Evidence / Results

The c36 evidence says the answer is not "all of it."

The quality branch:

• +16004 PnL
• 15 trades
• DSR 0.6400
• failed only trades_per_week_ok

The opportunity branch:

• 85 trades
• +2987 PnL
• lower-quality shape than the selective branch

This is a clean result because the tradeoff is not hidden. The repo does not need to argue that the quality branch was better. The numbers already show that the denser version did not preserve the same performance profile.

What Worked

What worked was selectivity. The high-quality version isolated a real signal. That is why c36 remains part of the current portfolio conversation rather than being closed with the failure-week branches.

The branch also worked as a research instrument. It taught the repo something very specific about intraday mean-reversion options: density is not a free improvement. Raising trade count by taking less extreme setups changes the branch economically and statistically.

What Failed

What failed was the attempt to solve the portfolio problem by loosening the setup. The extra trades did not carry enough of the original edge. This is a common research trap. The branch feels closer to deployable because it is more active, yet the change that made it more active also reduced what made it interesting.

There is an even harsher companion result in Episode 7. c37 tried to express related mean-reversion logic through 2-5DTE vertical debit spreads instead of through the 0-2DTE single-leg structure used by c36. On SPY, it produced 0 trades. That result shows that the monetization layer can extinguish the sample entirely, fully erase it.

Takeaway

Signal quality drops when you chase density because the extra trades are usually less extreme, less clean, and more sensitive to option-execution friction. The c36 and c37 branches make that lesson concrete. One branch stayed selective and sparse. The other attempts at widening or restructuring the opportunity set did not preserve the same value.

If you want the broader c36 case study, VWAP Mean Reversion Backtest: The Logic, the Edge, and the Failure Modes is the right companion. For the decision-level summary, VWAP Z-Score Strategy: How We Evaluated c36 and Why It Still Was Not Promoted explains why the branch remained below the promotion line. Join the research log to get the next backtest and failure report.

How the terminology applies

For Intraday Mean Reversion Options: Why Signal Quality Drops When You Chase Density, the options data workflow should treat Option chain snapshot, Contract snapshot, Volume/OI pressure, Options flow false positive, Scanner artifact, and Historical REST window as operational state rather than glossary decoration. That framing keeps chain selection, contract snapshots, activity filters, quote state, and endpoint access tied to the exact contract the page is discussing.

A developer implementing this Case Study idea should persist Backfill, DTE bucket, Moneyness band, Quote condition, Trade condition, and IV skew beside the result, instead of leaving those words in a term card. It also makes false positives easier to diagnose because a high-activity contract can be separated from a tradable, timestamped, and entitled data object.

The review artifact for Intraday Mean Reversion Options: Why Signal Quality Drops When You Chase Density becomes more useful when 0DTE contract, OCC root, Options data API, OPRA-originating data, OCC option symbol, and Bid/ask spread appear in the same body of evidence as the selected rows. When the article moves from concept to implementation, these fields should shape request order, cache boundaries, row labels, and review tables.

In production notes for this options data workflow, Midpoint, Quote/trade condition, Quote vs trade semantics, REST snapshot, WebSocket stream, and Entitlement gate define the checks that decide whether the workflow is reproducible. The result is a scanner or dashboard that explains why a contract was shown, skipped, refreshed, or passed into a downstream research step.

For Intraday Mean Reversion Options: Why Signal Quality Drops When You Chase Density, the practical acceptance test is simple: another developer should be able to read the body, identify the exact inputs, reproduce the request sequence, and explain the accepted and rejected rows without relying on the bottom terminology grid. If a phrase appears in the page vocabulary, it should correspond to a stored field, a validation check, a replay step, or an implementation decision in the options data workflow.

This is also the reason the article should not measure success only by the final chart, table, or headline metric. The better standard is whether the data path, timing model, entitlement state, and evidence trail survive review. When those pieces are written directly into the body, the terminology becomes part of the workflow readers can implement.

Density should be measured with quote state

Signal density is not just a count of more setups. In options, each additional setup brings a new contract snapshot, quote window, trade window, and reject path. The comparison should store VWAP deviation, z-score, selected OCC option symbol, DTE bucket, bid, ask, spread percent, quote freshness, implied volatility, Greeks, and open interest beside the trade count.

That makes the quality drop easier to diagnose. A looser entry rule may add setups that pass the stock signal but fail the option market: stale NBBO, no bid, thin top-of-book size, or quote conditions that block the fill model. The strategy did not merely become busier. It changed the mix of executable and non-executable rows.