The broad FLEX short straddle loses money after spreads and a one-day delay. It is the paper's only tradable result, a one-day trade sorted on implied minus realized volatility. The authors report the loss themselves. They also find an 11-trade subset that "remains positive even with latency", while declining to call it "a production-ready strategy".

FLEX prints and the regular chain

FLEX options are exchange-cleared contracts whose strike, expiry, exercise style and settlement can be negotiated. Cboe reports roughly 60,000 contracts a day in 2014 and nearly 800,000 in 2024. Annual growth was 26%, compared with 7.7% for listed options overall.

In FOMO: FLEX Options and Market Outcomes, Prakash, Ferstler and Pezolano examine whether single-name FLEX activity leaves a mark on regular listed options and the underlying stock. OCC, Cboe and Newmark Risk supply data for 250 trading days, from March 31, 2025 to March 27, 2026. The authors exclude indices, ETFs/ETNs and S&P 500 constituents (632 names), reasoning that those markets are too deep for a daily event test. The remaining universe has 910 listed underlyings; 41 show FLEX activity. Those 41 yield 68 consolidation rows across 53 event dates. Each event pairs an underlying with a FLEX effective date. Most have one consolidated FLEX row (mean 1.283 rows), and median time to expiry is 57 days.

The proposed mechanism starts with selection: institutions seek custom exposure in stocks with high listed IV. A FLEX print could therefore flag expensive volatility that a seller can collect. The abstract reports a matched-control difference-in-differences coefficient of 0.3061 on log listed open interest (BH-adjusted p 0.0168). Spread and IV are insignificant; volume has p 0.0724. For FLEX names in the mid-price tier, the ATM IV premium is 0.1235 (p 0.0030) after date fixed effects. The same-day abnormal return gap is -0.8547% (p 0.0442), then disappears by day +1 (0.2820%, p 0.5262). Finally, event-day IV minus realized volatility predicts execution-stressed short-straddle P&L with a slope of 0.0366 (p 0.0100). That calculation proxies the one-day move as 0.8×IV×sqrt(1/252) and compares it with the absolute next-day stock move.

The authors read FLEX as a marker of institutional volatility demand and option positioning, without a forecast for stock direction. I agree on direction. How much the marker contributes is less clear.

Where does the open-interest shift come from?

For the difference-in-differences test, the post indicator turns on at each name's first FLEX date and stays on through the rest of the year. The authors describe this as a sustained post-period level shift. Yet their windowed test finds nothing: across 53 events, log open interest changes -0.0389% (p 0.8883) when days 0 to +3 are compared with days -5 to -1.

Gradual inventory building could reconcile those results, although the paper does not address the paired null. In the authors' account, dealers facing a FLEX contract build offsetting listed positions over time. Open interest would then register accumulated positions rather than a one-day volume burst. Plausible. But a permanent post dummy also picks up names whose open interest was already rising when FLEX arrived. The pre-trend test has p 0.1997 and covers only the five days before the event, with tau -1 as its reference.

Contract quantity is absent for each row. The 0.3061 log-point shift (about 36%) therefore cannot be put alongside FLEX trade size. The paper cites Sun-Abraham and Callaway-Sant'Anna, though we did not find a Callaway-Sant'Anna or Sun-Abraham estimate in it. Event-time effects that kept growing well past day +3 would change my mind.

The familiar volatility spread

Across 51 FLEX events, regressing execution-stressed straddle P&L on event-day IV minus lagged 20-day realized vol produces a slope of 0.0366 (HC3 p 0.0100). Results climb through the quartiles. Q1 earns -0.9280% with a 46.15% win rate; Q4 earns 2.1068% with 76.92%.

This is Goyal and Saretto's volatility spread. The paper cites their work and says its result is consistent with it. The regression uses FLEX names alone, and we did not find an equivalent regression for matched non-FLEX names on the same dates. So the slope establishes that an existing signal works across 51 FLEX events; it leaves the print's incremental value unsettled. A matched non-FLEX IV-RV sort that underperformed the FLEX-conditioned sort would supply the missing evidence.

The selection result concerns IV levels, which leave that question open. Across tiers, the premium is 0.0481 (p 0.2708); in the low-price tier, it reverses to -0.1099 (p 0.1219). The high-price tier shows 0.2305 (p 0.0362), drawn from only 5 FLEX symbols. High IV and a wide IV-minus-RV spread are different measures. The straddle also takes its IV from the median of the whole chain, rather than from ATM quotes.

What survives a one-day wait?

For 40 rich-IV events, the research proxy earns 1.5350% per trade, with an 80.0% win rate and p 0.0031. The authors then charge 1.25 times the median chain relative spread, plus $0.65 per contract on four fills. Mean return falls to 0.4704% (p 0.3589). A one-day delay takes it to -0.4689% (p 0.4292), alongside a 52.5% win rate and a -28.72% drawdown. The delayed broad trade is a coin flip at 52.5%.

The delayed variant retains the event-day credit and moves only the realized stock move to the following day. The authors call it "intentionally conservative". The move gets the harsher treatment; the credit does not. A late entrant would sell at day-1 prices, which this calculation never charges.

An 11-trade subset, top IV-RV with tight spreads, stays positive. It earns 2.6144% after costs (p 0.0400) and 2.0737% with latency (p 0.0306). The authors present this as an illustrative double-conditioned subsample requiring out-of-sample confirmation. Their caution fits the construction: the liquidity cutoff comes from full-sample medians over the study window, bringing in spread information recorded after many selected events. Its reported Sharpe of 2.36 is mean over standard deviation times sqrt(11). Both that figure and the positive delayed result rest on 11 trades chosen using the full-sample liquidity cutoff.

We could not run this ourselves. Our platform lacks FLEX consolidation, contract and position data, leaving us unable to identify the events. The 2025 to 2026 window also extends beyond our historical coverage.

A different mix in index products

The evidence I had not seen before comes from the index and ETF products excluded from the main test. Across 2,267 product-days, FLEX put share is 19.7 points below same-day listed put share. Daily co-movement is weak (Spearman 0.098). PM-settled SPXW FLEX flow is 23.0% puts, versus 61.6% in the listed SPX chain; NDXP and RUTW register 6.9% and 6.6%.

FLEX flow carries a different mix of payoff demand than listed flow.