There is a paper I wanted to test, Gao, Xing and Zhang (2018), “Anticipating Uncertainty: Straddles around Earnings Announcements”, it says if you buy an at-the-money straddle (a call and a put at the same strike, you win if the stock moves a lot either way, or if the options just get more expensive) 3 days before a company reports earnings, you make +3.34% on average. The idea behind it is simple, traders pay more and more for options as the report gets closer, so you buy early and sell to them before the news comes out, you do not even need to guess the earnings.
It sounds good, but the paper used daily closing prices, and everybody who traded options around earnings knows the bid/ask spread gets wider at exactly that time. So the question I care about is not “is there an edge”, it is “is there anything left after I pay the spread”.
QuantConnect has what this needs for free, report dates with before-market or after-market timing (EODHD Upcoming Earnings), and minute bid/ask quotes on single stock options from 2012 (AlgoSeek US Equity Options). So I can buy at the ask and sell at the bid, which is the part the paper could not measure.
The rules, written down before any result
I wrote the rules and the pass rule in a file and committed it before the first P&L run, because I have tested enough ideas now to know what happens when I pick the rules after seeing the numbers.
- Top 100 US stocks by dollar volume, rebuilt every month, no ETFs
- Buy the straddle 15 minutes before the close, 3 sessions before the last close before the report
- Strike nearest the stock price, first expiry on or after the day the stock reacts
- Exit A (the main test), sell 15 minutes before the last close before the report, so I am out before the news
- Exit B, sell one session later, so I hold through the news
- Buy at ask, sell at bid, $1 per contract per leg, each way ($4 a trade)
- Skip it if the straddle spread is more than 10% of the mid price at entry
- 2012 to 2021 to look at freely, 2022 to July 2026 kept aside to run once
- It passes only if exit A makes money in the first period, and still makes money in the second one at the unlucky end of a resample
What came back
The short answer is, it fails, and it fails already in the first period, so I did not run the second one. ❌
| Exit A, out before the report | Exit B, held through the report | |
|---|---|---|
| Trades | 2,958 | 2,658 |
| Win rate | 31% | 29% |
| Average win / average loss | +$136 / −$128 | +$511 / −$392 |
| Average per trade | −$45 | −$130 |
| Return on the price paid | −3.84% | −14.90% |
| Worst trade | −$7,304 | −$17,044 |
Exit A lost money in 9 of the 10 years, the only good one is 2015 and that is +$10 a trade, which is nothing. Exit B is much worse, holding a straddle through the report loses about 15% of what you paid on average, this is the well known one (the options are priced for a big move and most of the time the move is smaller).
The edge is there, it is just smaller than the cost
This is the part I find interesting. If I ignore the spread and use mid prices for exit A,
- Return at mid prices, +1.79% on average (the paper says +3.34%, same direction, smaller)
- The median is −1.98%, so it is a few big winners carrying the average, not most trades winning
- Straddle spread at entry, 3.5% of the mid price
- Straddle spread at exit, 4.8% of the mid price
- After paying both, −3.84%
So the paper is not wrong, the options do get more expensive into the report. But I pay half the spread going in and half going out, about 4% together, and the edge is 1.79%. And these are the 100 most traded stocks in the US, the spread will not be better anywhere else.
Selling it does not work either
I also wrote down the other side, only to look at, with no pass rule. Sell the straddle right before the report and buy it back at 10:00 the next morning, 2,848 events.
- The options priced a 5.44% move on average, the stock moved 3.48%
- The real move was smaller than the priced move 76% of the time
- Win rate 53%, average −$28 a trade, worst trade −$20,427
Three times out of four the options are too expensive, and selling them still loses after costs, because the one time out of four is big, and the spread is there on this side too. A naked straddle has no limit on the loss, a real version needs wings (buying options further out to cap it), and this number tells me that one is not going to be easy either.
The calendar is not as clean as it looks
Before trusting any of this I checked the dates, for every event, is the biggest daily move in the 7 sessions around it on the day I expect the stock to react?
- Before-market reports, 43.5% of 2,303
- After-market reports, 61.2% of 1,131
- 19.3% of the before-market ones made their biggest move one session late
I expected something like 95% and that was too much to ask, a big stock often moves less on its own report than on a bad day for the whole market. But one in five moving a day late looks like some after-market reports are marked as before-market. For exit A it does not hurt (I am out before the report either way), for anything that holds through the report it matters.
One more thing I found on the way, the calendar has an “estimate” field, which could save paying for analyst estimates. For AAPL’s October 2019 report it says 0.71, in a quarter where AAPL earned about $3 a share. AAPL split 4 for 1 in August 2020, so that number was divided afterwards, it is not what somebody saw in 2019. The estimate also changed between rows in 46 of 3,592 events. So I would be careful before building an earnings surprise on it.
What is left
I need to be honest about the limits, it is one test with one set of rules, 1 straddle a trade so the dollar numbers are led by the expensive stocks (the percent numbers say the same thing), and I checked only four trades by hand (JNJ, GS, UNH and JPM, all reporting on 15 October 2019).
The 2022 to 2026 data is still untouched. If I come back to this, the only version that makes sense to me is the stocks with the tightest option spreads, maybe 20 names, where the cost could be under 1.79%. That is a new guess, so it needs its own rules written down first, and I have not decided if it is worth the time, the median trade loses even with no cost at all.