Selling Earnings Volatility Strategy
Understanding Edge
Without edge, no matter what strategy you choose, you will not be profitable in the long run.
What is Edge?
Edge is a repeatable, statistical advantage, that gives you a positive expectancy over time. It's not about winning every trade, it's about having a systematic process, that when executed consistently, results in profitability over a large number of trades.
Many traders confuse edge with luck or random patterns in data. Just because it worked a few times in back-tests, does not mean it's an edge. You need a valid, logical reason for why the edge exists. Otherwise you're just data mining and eventually the market will correct itself, leaving you with nothing.
To summarize so far:
- People assume because something has worked a few times that they have found an edge
- We need a valid logical reason for the edge to exist:
- If you can't explain your edge simply in a few minutes its probably not a good one
- An indicator or model is not an edge; edge comes from understanding why someone is willing to take the other side of your trade
- Remember for you to make money, someone else has to lose money (why are they paying you?)
- If you don't have an edge the best long term outcome you can hope for is to break even
Why can I monetize edge?
This brings up a critical question: If edge is real, why hasn't it been arbitraged away by larger players?
Some institutions have billions of dollars and entire teams of quants searching for inefficiencies, so how can we, as individual traders, still monetize this?
The answer lies in understanding where large players don't want to take risks. The best trade ideas come from areas where other market participants are offloading risk at any cost. They're not trying to optimize for maximum returns, they're simply looking for certainty; and when people are desperate for certainty, they overpay for protection, creating inefficiencies we can exploit.
However, every trade idea must be testable with data. If you can't quantify it, you're just gambling. Too many traders rely on gut feeling and intuition. But if you don't have a way to falsify bad ideas, then all you're doing is trading on hope, and hope is not a strategy.
Even when you find an edge, you have to accept that edges are noisy. They don't work every time, and results can vary in the short term. This is where most traders fail, they abandon a good strategy during a draw-down, because they don't understand the variance of their strategy; this is why proper research is so important. If you don't have the data to support your strategy, then how will you know if you should keep trading through a draw-down, or if your edge has disappeared? Later in this page we will go over the data that answers this exact question.
Proposed Strategy
So what's the strategy then?
Selling earnings volatility. That's it!
It sounds simple, but there's a lot of nuance that we will address, with the data later in this page.
This works because quarterly earnings events often account for the majority of the volatility that a stock will experience in the entire year. By selling options ahead of earnings, we aim to profit from 2 key factors.
First, what is commonly referred to as "implied volatility crush" (or IV crush), which is the rapid drop in implied volatility after earnings are announced.
The second is stocks moving less than expected. Our idea is that options tend to overprice earnings moves, and when the stock doesn't move as much as anticipated, we profit from the discrepancy.
So why does this edge exist? Because people hate uncertainty. Especially big institutions, and funds with tight risk controls; and uncertainty in a stock is directly reflected in implied volatility (IV). Around earnings, IV spikes because traders, funds and retail investors rush to buy protection. Many of these participants aren't concerned whether IV is over or under-priced, they just want to hedge their positions. We can call them "price-insensitive traders" and they create an opportunity for us.
Beyond hedgers we also have speculators, who are traders looking to gamble on earnings moves. They often buy short-term upside call options. This increased demand pushes options prices and volatility even higher, further fuelling our edge.
When you combine:
- Hedgers willing to overpay for protection
- Speculators driving up demand for options
- General uncertainty leading to higher implied volatility
You have a market imbalance where options are systematically overpriced, leading into earnings, and we can take the other side of that trade!
How can we monetize this strategy?
There are multiple ways to structure short volatility trades around earnings. But we will focus on two:
- The short straddle
- The long calendar spread
Short Straddle
The short straddle is the most common short volatility strategy. It involves selling one call and one put of the same strike at the same expiration. Generally, this will be ATM (at-the-money) strike price. This means that we profit when the stock moves less than the expected earnings move and when implied volatility collapses post-earnings.
However, because we are selling short-term options the gamma risk is very high. Gamma measures how fast delta changes, and near earnings with inflated IV, and short time to expiration gamma is extremely sensitive. If the stock moves more than expected our vega gains from IV crush won't be enough to offset the loss from gamma, and we can take a big hit. (1)
-
The Greeks are risk measures that describe how an option's price changes in response to various factors:
-
Delta (\(\Delta\)): Measures the rate of change in option price relative to the underlying stock price. Ranges from 0 to 1 for calls (-1 to 0 for puts). A delta of 0.50 means the option price moves $0.50 for every $1 move in the stock.
-
Gamma (\(\Gamma\)): Measures the rate of change of delta. High gamma means delta changes quickly as the stock moves, creating more risk and opportunity. Gamma is highest for ATM options near expiration.
-
Theta (\(\Theta\)): Measures time decay – how much value the option loses each day as expiration approaches. Negative for long options, positive for short options. Accelerates as expiration nears.
-
Vega (\(\nu\)): Measures sensitivity to implied volatility changes. Higher vega means the option price is more sensitive to IV changes. Long options have positive vega (benefit from IV increase), short options have negative vega (benefit from IV decrease).
-
Rho (\(\rho\)): Measures sensitivity to interest rate changes. Usually the least important Greek for short-term trades.
-
Long Calendar Spread
The long calendar spread involves selling a near-term (a.k.a. front-month) call option and buying a longer term (a.k.a. back month) call option, at the same strike. It is usually executed with the shared strike, being the ATM strike. It is a debit strategy, because the option we are buying at the back will be more expensive in dollar terms, than the one in the front. Typically, we will use a 30 day expiration gap because it maximizes exposure to earnings volatility, while keeping the structure stable.
Example
- SELL Feb 7 $200 Call for $2 Credit (vega = -15)
- BUY Mar 7 $200 Call for $3.50 Debit (vega = +35)
- Total Position Debit: $1.50 (equal to max. loss)
A common point where people get confused: A calendar is technically long vega, but this is misleading. The back-month option has more vega than the front-month, making it seem like we want IV to rise. But in reality, what we want is IV to drop in the front month, more than it does in the back month (by a factor of roughly 2.3x to break even). Essentially we want our vega gains in the front contract to offset our vega loss in the back.
We are also short gamma, because the option we sold in the front has a higher negative gamma than the positive gamma in the option we bought in the back.
This means that we want the stock to move as little as possible, making it so that our gamma losses in the front are as small as possible, since the back month positive gamma will not be able to compensate.
Lowering front-month IV raises our profitability, but if back-month IV also drops significantly, our structure collapses inwards, limiting profits. Through testing we found that a 30 day difference minimizes this issue.
Choosing Between the 2 Structures
So which one is better? The straddle offers higher potential returns and lower commissions, but also has higher tail risk, meaning that when things go wrong, they go very wrong. The calendar offers better risk control and a smoother equity curve, but it generally has lower returns.
In the data section, we'll analyze thousands of historical events to determine which structure performs best over time.
Dataset
Now let's dive into the numbers. Everything we show in this section is backed by real market data; no cherry-picking, just raw numbers and analysis. We're working with a dataset of 4500 unique stocks spanning from 2007 to 2019, covering a total of 72'500 earnings events.
Within the dataset, we track several key metrics: both predictor variables (which help us determine when to enter a trade) and target variables, which measure the success of each strategy.
Predictor Variables
Term Structure Slopes
This measures the steepness of the implied volatility term structure. Specifically, it looks at the difference of IVs between the front month (near term expiration) and further expirations (45 days or higher). A steeper structure will show that near-term options expect large moves, but they expect volatility levels to fall back to normal levels over time.
There's also the term structure ratios: similar to the term structure slope, but expressed as a ratio rather than a slope.
30-day Average Volume
We use the 30 day average volume, which we use as our liquidity measure. This will help us determine if liquidity plays a role in pricing earnings events. Based on our initial hypothesis, we should see returns increase with volume. More volume means more activity, which means more hedging, speculation, etc.
IV/RV Ratios
Finally there are IV/RV ratios, which compare implied volatility to realised volatility over the period prior to earnings. This is used to see if there is predictability coming from IV being overpriced, in the time before earnings.
Target Variables
We analyse 3 primary return-based metrics. Note that all of these positions were opened 15 minutes before the close of the trading period prior to the earnings event.
The first (what we call "Straddle Return Jump") is the return of the trade closed 15 minutes into the trading session after earnings. We also have "Straddle Return Move" which is the return is we close the position 15 minutes before market close on the day after earnings. Finally, we also have the "Calendar Return Jump" which is the return of the calendar spread closed 15 minutes into the trading session after earnings.
All of these include realistic trading costs, including Interactive Brokers commissions and slippage estimates based on bid-ask spreads and volume. This ensures that everything we're analyzing reflects real-world trading conditions, not just theoretical models.
Results
Results of Straddle Jump Strategy
When we plot the historical returns of the Straddle Jump strategy (graph below) we immediately see something striking. Most returns cluster around small profits, but there's a long, dangerous left tail where rare, but massive losses occur.
In fact, as shown in the tables below, 1% of the time, this strategy lost 130% or more and 0.1% of the time it lost over 410%.
Distribution Statistics
| Statistic | Value |
|---|---|
| Count | 72426 |
| Mean | 0.000032 |
| Std. | 0.632031 |
| Min. | -92.165025 |
| 25% | -0.103015 |
| 50% | 0.038753 |
| 75% | 0.193379 |
| Max. | 0.918400 |
Tail Risk Analysis
| Occurence | Gain/Loss |
|---|---|
| 5% of time | -0.5915 |
| 1% of time | -0.4212 |
| 0.1% of time | -4.2204 |
| 0.01% of time | -16.7656 |
We also notice the near-zero percent mean return, indicating that trading all events blindly, will just barely breakeven over time. This is generally expected and shows that on average the market prices earnings events quite well.
Results of the Calendar Jump Strategy
The calendar's distribution looks far more stable with lots of small consistent winners, and fewer extreme losses. The worst case scenario for the calendar is losing the entire debit paid. We also notice the near-zero mean percent return here showing that we would break even if we just blindly traded all events.
| Statistic | Value |
|---|---|
| Count | 67740 |
| Mean | 0.000006 |
| Std. | 0.207757 |
| Min. | -1.071140 |
| 25% | -0.083854 |
| 50% | 0.008669 |
| 75% | 0.110970 |
| Max. | 0.788395 |
Comparing Strategies Quarter-by-Quarter
The Straddle jump shows high fluctuation around the zero line, showing no consistent long-term edge from trading blindly. The calendar spread also showed fluctuations around the zero line, showing that is trading all events, there is no real long term edge there. This is why we have predictor variables: to see if we can improve this fact.
One interesting observation: The move straddle that was held until close the next day was a consistent loser, affirming the well-documented, post-earnings announcement drift effect.
PEAD
Post-earnings announcement drift (PEAD) is a well-documented market phenomenon that results in prices reacting to earnings announcements slowly over time. For example, if a stock jumped 5% in the morning after earnings, it is more likely to continue to rise throughout the day. This means you can sell the position in the morning after the jump, and take the opposite position to ride the predictable change. Because the jump shows more promise here, we will only consider that for the rest of this study.
What do our predictor variables predict?
So do any of our factors actually predict profitable trades? When we split the data into deciles (10 equal groups), based on each predictor variable, we found 3 factors that correlated strongly with success:
- The term structure slope, specifically the nearest expiration to the 45 day expiration. The more negative the slope the higher the returns for both the straddle and the calendar → The steeper the IV curve, the more overpriced short-term options tend to be. This shape of the slope is commonly referred to as backwardation, meaning that the near-term options are more overpriced in IV terms, than later term ones. The market expects volatility levels to drop back to normal after the event.
- The 30-day average volume. Higher pre-earnings trading volume lead to better returns for both the straddle and the calendar. This suggests that more volume (i.e. more participants) will lead to a higher level of price insensitivity where demand is more likely to outpace supply
- The IV30/RV ratio. The higher this ratio, the better the expected return. This confirms that if IV was overpriced in normal conditions it was even more likely to be overpriced for earnings.
The Model
With these predictor variables known, we built a simple rule-based model that only trades when:
- Term-structure slope is sufficiently negative (at or below 5th decile, 1-indexed)
- The 30 day average volume is above a key threshold (6th decile)
- The IV30/RV ratio (at or above 5th decile)
These rules filtered out 88% of events for the straddle and 90% of events for the calendar, leaving us with only the highest quality trades.
The final results of the straddle show that average return became +9% but with a high variance, and the calendar mean return got up to 7%, with a lower variance.
Straddle Strategy Results
| Statistic | Value |
|---|---|
| Count | 8945 |
| Mean | 0.090433 |
| Std. | 0.484275 |
| Min. | -8.386427 |
| 25% | -0.085909 |
| 50% | 0.174166 |
| 75% | 0.391289 |
| Max. | 0.918400 |
Calendar Strategy Results
| Statistic | Value |
|---|---|
| Count | 7313 |
| Mean | 0.072829 |
| Std. | 0.286135 |
| Min. | -1.071140 |
| 25% | -0.062751 |
| 50% | 0.111473 |
| 75% | 0.264228 |
| Max. | 0.788395 |
Now:
- Both mean returns sufficiently positive to establish edge
- Straddle has a higher return but also higher variance when compared to calendar
- The calendar was much more stable, making it easier to trade consistently
This is a good backlog of data to trade edge, but we still need more to trade. We still need to test how this strategy would perform in several scenarios including what position size to use. We will also need strategy metrics, such as max. draw-down, win-rate, etc.
To help us define this, let's put our strategy through a Monte-Carlo simulation.
Simulation Results
Kelly Criterion
The Kelly criterion determines the optimal fraction of your portfolio to allocate to each trade. "Full Kelly" is the peak of this distribution usually denoted \(f*\) (\(K\) on the graph). \(y\)-axis is the growth rate, \(x\)-axis is the fraction of the portfolio used on each bet. Theoretically, Kelly-sizing maximizes long-term growth, but in practice, full Kelly is almost always too aggressive, leading to extreme volatility, and an uncomfortably high risk of bankruptcy.
- For the straddle, the suggested Kelly fraction suggested is 6.5% (\(f* = 0.065\)) (e.g. if we had a 10'000 account, it recommends selling a straddle with a maximum of 650 in collected premiums)
- For the calendar, the Kelly fraction is 60% (\(f* = 0.6\)) (e.g. for the 10'000 account the recommended position size would be 6'000 debit)
By using the Kelly fraction as a proxy for the better strategy, this suggests that the calendar structure is likely superior (the straddle has a much lower Kelly fraction due to its extreme left tail).
However, even though the calendar losses are not as high as with the straddle structure, a 60% fraction is far too high, and we'll show exactly why this is in a moment.
Monte Carlo Simulation
Now we shouldn't just rely on historical data, we will run Monte Carlo simulation to get a better picture of real world variance. Backtests only show one historical path, while Monte Carlo runs thousands of possible outcomes by randomly sampling from the dataset. This gives us a much better idea of what can happen in the future and whther our strategy holds out in multiple market conditions.
For this study, we ran 10'000 simulated PnL paths over a span of 1'000 trading days (roughly 4 years) and another batch over 252 trading days (1 year), all starting with a $10'000 portfolio. The account is marked as "bankrupt" if the portfolio drops below $500.
Full Kelly - Straddle Structure
Running the simulation for a full Kelly fraction in the straddle structure, we see no bankruptcies and some paths manage to hit returns from 35k to over 4M in the same period. This looks great right?
| Statistic | Value |
|---|---|
| Count | 10'000 |
| Mean | 1'638'276.63 |
| Std. | 802'975.62 |
| Min. | 35'370.55 |
| 25% | 1'033'445.63 |
| 50% | 1'635'279.17 |
| 75% | 2'205'342.24 |
| Max. | 4'589'872.58 |
Not so fast. When we examine the draw-down histogram, we can see a big problem. About 35% of all paths experience a maximum draw-down of over 45%, with some paths dropping as low as 80%. This means that at some point over the 4 year period, more than a third of traders will have seen half of their capital wiped out, or worse. If we look at the longest draw-down duration, most of them lasted between 100 and 200 trading days, with some going all the way up to 600 days (2.5 years)!
Now ask yourself, can you see your account drop by 45% or more? Can you keep executing your strategy even after losing 80% of your capital? This is why trading size is important. The variance is real, and if you don't control your risk, you won't survive long enough to see the gains.
Full Kelly - Calendar Structure
Looking at the results for the full Kelly calendar structure (60% bet per trade), we see that roughly 5% of cases end up in bankruptcy.
| Statistic | Value |
|---|---|
| Count | 10'000 |
| Mean | 3'315'134.30 |
| Std. | 873'702.33 |
| Min. | 429.39 |
| 25% | 3'098'258.29 |
| 50% | 3'455'859.11 |
| 75% | 3'788'282.62 |
| Max. | 5'169'410.42 |
Checking the draw-down histogram, we see an even more troubling trend: Many of the drawdowns are in the 80-95% range (yikes), and the distributions of draw-down is skewed heavily towards 100%. This guarantees eventual blow-ups.
This tells us that betting at full Kelly has way too much variance. However thanks to this variance, this strategy was able to turn 10'000 into 1M in 1 year. To reduce the variance, let's use fractional Kelly going forward.
Fractional Kelly
Fractional kelly bets let us reduce our variance, while not reducing our returns by as much (e.g. trading at 50% Kelly reduces variance by 50%, but only reduces returns by 25%). We can use this knowledge to trade smaller reducing our variance while preserving our return, so let's scale back to something more reasonable. Let's try a 30% Kelly bet.
Fractional Kelly - Straddle Structure
For the straddle, this means dropping down to 2% bet per trade, and over the course of the 4 years simulation period, the average returns drops, however the maximum draw-down also drops significantly.
| Statistic | Value |
|---|---|
| Count | 10'000 |
| Mean | 61'836.60 |
| Std. | 19'234.80 |
| Min. | 14'802.44 |
| 25% | 47'922.36 |
| 50% | 59'415.96 |
| 75% | 73'153.41 |
| Max. | 178'658.83 |
The largest draw-down is now about 37%, and the average draw-down is only 15% (much more manageable). The mean Sharpe ratio over the 4 years is around 3%.
Fractional Kelly - Calendar Structure
For the calendar structure, we take the 30% Kelly bet with translate to 18% per trade. This return graph looks very strong, however the drawdowns are still uncomfortably high.
| Statistic | Value |
|---|---|
| Count | 10'000 |
| Mean | 2'914'522.41 |
| Std. | 476'353.99 |
| Min. | 918'294.34 |
| 25% | 2'599'679.78 |
| 50% | 2'923'982.73 |
| 75% | 3'231'677.65 |
| Max. | 4'618'254.53 |
The maximum draw-down is about 76% with the mean around 40% which is still too high.
So let's reduce to 10% Kelly (6% per trade) to further decrease this risk. This finally brought the draw-downs into a comfortable area.
| Statistic | Value |
|---|---|
| Count | 10'000 |
| Mean | 771'913.23 |
| Std. | 368'506.65 |
| Min. | 84'125.80 |
| 25% | 486'817.49 |
| 50% | 705'800.62 |
| 75% | 1'004'502.75 |
| Max. | 2'378'622.44 |
The returns still look great for a starting amount of this size, and the strategy looks much more sustainable than before.
At this level, the Sharpe ratio was actually higher than the straddle, making this a preferable structure to trade.
Which structure is better?
Both structures are viable, however we see multiple indicators throughout the study showing that the calendar structure is a better risk-reward balance. Multiple earnings events which create higher-tan expected moves where the straddle structure would result in a devastating loss.
Taking a Longer-Term View
Now that we have selected the 10% Kelly calendar structure, let us take a look at the PnL over 10 years. Starting from 10'000, the mean ending portfolio value is around 6M (compound annual growth rate of 90%). The max. draw-down distribution sees a mean max. draw-down of ~20%, and mean longest draw-down of around 6 months. The win-rate is 66%, the expectancy per trade is 0.265%, and the mean sharpe ratio is 3.5%.
| Statistic | Value |
|---|---|
| Count | 10'000 |
| Mean | 6'185'119.72 |
| Std. | 726'647.62 |
| Min. | 3'352'922.71 |
| 25% | 5'701'554.64 |
| 50% | 6'184'604.87 |
| 75% | 6'675'520.85 |
| Max. | 8'678'266.56 |
Overall, this strategy performs extremely well, but we said earlier, position sizing is everything. No matter how strong an edge is, if you size too aggressively you will blow up. Risk management is the key to longevity and preserving capital is always the #1 priority.