HV is chunky around earnings so you cannot RV against implied long term as you're adding events each quarter, hence the massive dev vs the marketable straddle as the events are embedded in the straddle. I don't understand what you're trying to accomplish. If you arrive that the straddle should be sold due to IV/HV>1 then you're going to be short vol (short combo) through reports or long term structure through reports (long cals). I would not approach vol-trading with such a method in equities. You'd also need to norm against index vols (HV and IV, skew/smile). TLDR, your short term model will be short IV into events and you'll be a seller long term as the IV represents vol-events added each quarter reflected in IV. There is some backwardation beta to index (and vega-sens) so your longer term modeling will probably do better.
So your HV model is screaming buy (HV lull prior to earnings) and you need something actionable so you would either buy calendar structures or short verts (single exp; straddles, strangles). You're a buyer of the calendar due to term structure edge in IV (M1/M2>1) or short the combo due to IV/HV. Both generally suck and are the low info option. Low info in the sense that it ignores the biggest vol-generator (Q-earnings). Risk to dot shot in the short straddle and the risk to the vol-line post report.
You need to model the derivs, as HV is generally a dead end. Even if you model with confidence you'll be hesitant to trade it large barring that one hiccup. There are arb-conditions in term structure positions held through reports when the switch value (M1/M2>1) screen builds stressed to a post-event figure. No guessing. Model vol, model switch (M1/M2), model post event crush to both. If all areas (structure stress) under the curve remain above x then you're in an arb. The only variable is your post event vol-line and you have quarterly data on that figure. Say that CSCO M1 vols come in 20 handles post report (50 to 30-line). You model -20 globally and price the structure. If it still looks good then you can likely trade it to the limit of your net liq. Generally these are short calendar/complex structures. I can't go into screen build, structure detail.
One data point. NVDA in the pre DeepSeek days had a 10X return on debit the day of earnings and another 10X post report.
Really insightful breakdown .. I like how you're using EMA-based MAD instead of variance. It's cleaner when the goal is option pricing or straddle comparison without distortion from tail events. That shortcut (MAD ≈ 0.8 × StdDev) works reasonably well, though like you said, it's not always consistent especially with skewed or jumpy assets. Your 365DTE SPY vs market straddle comparison is a sharp way to highlight relative overpricing especially when your modeled $79 vs markets $84 still keeps in line. But that $287 for NFLX vs market's double… that's a real outlier. I wonder if that’s implied vol expansion due to catalyst pricing (earnings, macro, etc.) or just misalignment in volatility surface assumptions. On the percentiles: comparing to ROC’s 16th/84th does have autocorrelation bias.. but I’ve found some value using rolling empirical vol cones layered with ranked MAD/StdDev ratios. Basically to separate true fat tails from trending expansion. I’m also experimenting with fuzzy regime classifiers where implied vs historical spread, percentile skew, and vol clustering behavior form rule based zones for whether to buy straddles, delta hedge, or fade gamma. Your approach already sounds robust even without hedging if you add dynamic delta or relative vol triggers it might outperform the raw odds you mentioned. Let’s keep iterating this kind of open volatility framework is where real edge forms
I love everyone. And it hurts me dearly when I have to drop the hammer on someone who's being a dips*%t.