All Activity

So I tested it. What was measured Every completed pre-earnings cycle in the backtest history, restricted to long straddles and strangles: 31,027 of them, once the test conditions below are applied. For each cycle, I computed the entry cost as a fraction of spot, then ranked it against the same figure on previous cycles of the same setup — same ticker, same strategy, same buy day. That gives a percentile: today's entry is cheaper than X% of the cycles that came before it. Then I sorted every cycle into five buckets by that percentile and looked at what each bucket returned. The part that matters: previous cycles only This is the detail that makes the test worth anything, and it is easy to get wrong. If I ranked each cycle against the full history — including cycles that had not happened yet at the time of the trade — I would be using information no trader could have had. Every result would look better than reality, and the error would be invisible. So the percentile for a cycle in March 2023 is computed only from cycles before March 2023. A setup enters the test only once it has enough prior history to rank against. That is what walk-forward means, and it is why the sample drops from 46,717 completed cycles to 31,027 testable ones. It is a costly constraint. It is also the only version of the test worth reporting. The result Entry percentile Cycles Win rate Median return Cheapest 20% 5,985 46.3% +4.5% 20–40% 5,932 44.5% +3.6% 40–60% 5,934 42.6% +1.4% 60–80% 5,938 41.3% +0.5% Richest 20% 7,238 40.9% 0.0% Read the columns downward. Win rate falls monotonically from the cheapest bucket to the most expensive. So does the median return, from +4.5% to zero. Expectancy behaves the same way: the cheapest quintile came out 2.1 percentage points better than the most expensive. Three different measures, roughly six thousand cycles each, all pointing the same direction with no reversal along the way. That is about as clean as this kind of test gets. It also says something plainer: on this data, the single most useful thing you can know about a pre-earnings volatility trade is what you are paying for it relative to what that same trade has cost before. The uncomfortable half I ran the same walk-forward test on the other filter people use — and on the one my own table sorts by. Filtering on historical win rate does exactly what it says: it raises the realized win rate, from 44.5% across all cycles to around 58% on the strictest setting. That part works. But expectancy moves the other way. Every threshold I tested — 60%, 70%, 75%, 80%, across three different history requirements — improved the win rate and degraded the expectancy. Twelve combinations, no exception. The mechanism is not mysterious once you see it. The exit rule caps the gain: the position closes on the first close at or above +10%, so a winner books roughly ten percent and no more. Nothing caps the loss. Filtering for setups that hit their target often selects positions that win small and often — and lose large when they miss. This is the same trap I wrote about in June under a different name. A high probability of profit is not an edge. It turns out the point applies to my own default sort. What happens when you combine them Cheapness and win rate are independent, and they pull in opposite directions. At any fixed win-rate threshold, cheap entries beat expensive ones by 2.4 to 2.6 percentage points of expectancy — consistently, at every level. And at any fixed cheapness bucket, raising the win-rate threshold lowers expectancy. Crossing the two does not rescue the second effect; the best bucket in the whole test is simply cheap entries with no win-rate filter at all. What I am doing about it The honest answer is: not yet decided, and I would rather say that than pretend. Sorting a table by win rate is what users expect, and the figure is not meaningless — it describes how often a setup has worked. But if the number people sort by moves expectancy the wrong way, the default is doing something I would not defend if asked to justify it from first principles. What is already true is that the entry-price block is not decoration. It measures the one thing in this data that improved every metric at once, and it sits on every card. Caveats worth stating The percentile needs history: a setup with four prior cycles produces a percentile that means very little, which is why the test requires a minimum before a cycle enters it. The buckets are wide. "Cheapest 20%" is not a threshold you can trade — it is a direction, measured across the whole population. And none of this says a cheap entry will work. It says that across thirty-one thousand cycles, cheaper entries returned more than expensive ones, on three measures at once, with no crossover. That is a tendency, not a promise, and the distinction matters more in this business than in most. OptionBench is a research and analysis tool, not an investment advisor. Nothing here is a recommendation to buy or sell any financial instrument. Backtested results are hypothetical and do not guarantee future performance. The original article was first published here.

Whether that is a lot depends entirely on what you compare it to — and there are three defensible comparisons, each answering something different. This is what the entry block on every OptionBench card is built around. Three reference points, three questions Position cost, against its own history. The same setup on the same ticker at the same point in the cycle has been priced many times before. Twenty past cycles, twenty entry costs, each divided by the spot at the time. Today's cost sits somewhere in that distribution. This answers: am I buying early or late in the ramp? The value of a pre-earnings long-volatility position comes largely from implied volatility rising as the announcement approaches. If today's cost already sits in the upper half of past cycles at the same T-x, much of that rise has happened. If it sits low, there is more room ahead. Implied volatility, against its own history. Closely related to the first, but not the same. The cost of a position depends on volatility, but also on time remaining and — for a long strangle — on how far apart the strikes sit. The two readings can diverge, and when they do it tells you something: a position that is expensive while its IV is ordinary is expensive for structural reasons, not because the market is bidding up volatility. This answers: how much ramp is left to build? Earnings premium, against what the stock delivers. The first two compare the ticker to itself. This one compares a price to a real counterpart: the ATM straddle divided by spot is what the market charges for the move, and the average past earnings move is what the stock has historically produced. This answers: what is the market valuing today? Why the third one is different The first two are self-referential. They will tell you a $30 straddle on a violent biotech is "cheap" if that stock usually prices even higher. That is useful — it means you are getting a better-than-usual price for that particular name — but it says nothing about whether the price is reasonable in absolute terms. The third comparison is the only one where the denominator is something real. On the BABA card above — captured on August 11, ten business days before the August 20 report — the market is asking for a ±8.8% move, and the stock has averaged ±5.6% across its last sixteen earnings. The ratio is 1.6. Measured across thirteen tickers inside their entry window on a single day, that ratio had a median of 1.72. Not 1.0. The market consistently charges more than the historical move — which is exactly what you would expect if the seller of volatility is being paid a risk premium. One caveat that matters: the straddle used sits on the chain that expires about a week after earnings, so it prices the announcement gap plus a handful of ordinary sessions. Part of the ratio is mechanical, which is why the reading is not centred on 1. Cheap against yourself is not cheap against reality Here is a real case from the scanner. BABA, twelve business days before the report. The entry cost sat below 56% of past cycles at that same point, and entry implied volatility was lower than 69% of them. Both gauges green. Nothing about this entry is expensive by the ticker's own standards. And yet: the market was pricing a ±8.9% move against a ±5.6% average earnings move over sixteen cycles. A ratio of 1.6. Both readings are true. The first says this entry is cheap for BABA. The second says you are still paying more than this stock typically delivers. Neither one is the answer. Holding both at once is the answer. What the Color does and does not mean On the card, the first two readings turn green when the entry is favorable to a volatility buyer and amber when it is not. The bar fills with distance from the median, so a full green bar is not good news by itself — it means this entry is cheaper than every cycle on record, which is worth knowing, not worth acting on alone. The third reading carries a number rather than a percentile, because the number is intuitive on its own: 1.6× the typical move means the market has priced in 160% of what the stock usually does. The color flags the extremes; the figure does the talking. None of this is a signal. It is context — the difference between a number on a screen and a decision. The reading that should actually stop you If you take one thing from this: the gauge that ought to give a volatility buyer pause is not the premium being high. High implied volatility on a long-volatility position is the price of admission to a wide distribution, not evidence of a bad trade. What should stop you is paying for a move the underlying does not historically deliver. That is the third reading, and it is the one most screeners never show — because it requires comparing what the market charges with what the stock has actually done, cycle after cycle. Next: why the position exits before the announcement, and what that changes about all of this. OptionBench is a research and analysis tool, not an investment advisor. Nothing here is a recommendation to buy or sell any financial instrument. Backtested results are hypothetical and do not guarantee future performance. The original article was first published here.