Skip to content

Why Do Automated Retail Option Orders Synchronize?

Bottom line: Brief SPX 0DTE trading bursts at round clock times are consistent with shared automation templates. Synthetic leg-based quotes can tighten while effective execution spreads widen. OPRA does not identify retail bot accounts, and the paper does not establish reliable strategy profits.

Source and version

  • The Rise of Algorithmic Retail Option Traders
  • SSRN author order: Pedro Angel Garcia-Ares, Diego Amaya, Neil D. Pearson and Aurelio Vasquez; the PDF cover lists Diego Amaya first
  • Official SSRN record · DOI
  • First posted on SSRN: April 17, 2026; revised online: August 13, 2026
  • Reviewed version: the 50-page PDF dated August 12, 2026, including tables and Internet Appendix; not independently replicated
  • Filed under 2026 Q2 by first SSRN posting. This is a current reading of the August revision, not information shown to have been available in April, nor a claim about earliest circulation elsewhere
  • Publication status: preprint; read denotes source review, not empirical validation

Question, data and identification

Can independent investors generate synchronized order flow by using similar platform templates and execution schedules? Does that clustering change trade composition and costs?

The authors use SpiderRock OPRA trades and quotes for Cboe C1 SPX/SPXW 0DTE, May 19, 2022–January 31, 2026. Main event comparisons use January 2024–January 2026, contrasting the first minute of half-hour windows from 10:00 to 15:30 with the preceding five minutes.

They reconstruct complex parent trades from matching timestamps and underlying prices, identify strategies from their legs, and infer trade direction from net premium relative to leg midpoints. These are reconstructed orders and inferred directions, not observed customer identities.

Findings

Clock-time bursts and standardized structures

Activity rises within seconds of scheduled times and usually fades within 30–60 seconds. The event windows contain more short call/put verticals and short iron condors. This fits retail automation defaults, but institutional algorithms could look similar. Account-level validation and direct platform-parameter matching remain future work.

A window classifier, not a return model

LightGBM uses 24 trade/market features without time-of-day. The first 80% trains the model, with five expanding-window validation folds, and the last 20% is held out. Table 7 reports holdout ROC AUC 0.6848, precision 0.52 and recall 0.61.

The label is whether an executed trade belongs to a burst window—not verified bot identity, future returns or fill probability. Some inputs describe the completed execution and therefore are not automatically available before an order.

This experiment compares the first 60 seconds with minutes 25–29 of a half-hour window, setting other trades aside. That reference differs from the preceding-five-minute comparator in the event tables.

Tighter quotes need not mean cheaper execution

Strategy Synthetic quoted half-spread, before → burst Effective half-spread, before → burst
Short call vertical 0.104 → 0.095 0.039 → 0.043
Short put vertical 0.107 → 0.102 0.039 → 0.044
Short iron condor 0.171 → 0.167 0.049 → 0.055

These Table 9 spread-price units are not complete per-contract or account costs. The quoted measure sums single-leg NBBO half-spreads; it is not an executable combo quote. Effective spreads use the package execution and midpoint benchmark.

Nor does quote tightening occur uniformly: synthetic quotes widen at 10:00, while pooled narrowing is concentrated later in the day. The evidence calls for time- and structure-specific execution analysis, not a universal clock-time rule.

Limits

  • Retail/bot identity and the common-platform mechanism are inferred, not directly identified or causally established.
  • Exact-time grouping can split genuine packages when legs print microseconds apart, understating some multi-leg structures.
  • Table 5 daily aggregate expiration profits are imprecisely estimated. They do not establish dependable profitability, safety or normalized risk-adjusted returns.
  • Overall complex-trade shares differ between Tables 1 and 2 without a reconciled denominator. We do not combine those percentages.
  • The historical SPX setting does not establish effects in other products, venues or present-day accounts. No independent data/code replication was performed.

Research implications

Treat these as hypotheses and diagnostics, not validated production changes:

  1. Stratify execution-cost analysis around clock times while controlling for date, volatility, strategy and order size.
  2. Preserve parent strategies, constituent legs and actual net execution premiums; keep executable package prices separate from synthetic NBBO benchmarks.
  3. Test any delayed or randomized scheduling against waiting costs, missed trades, price movement and fill probability. The paper does not prove such scheduling improves net returns.
  4. Separate window/structure classification from return prediction and deployment decisions.

Research boundary

This is original editorial analysis of a preprint, not replicated evidence or a basis for automatic trading deployment.