As you may already know, albeit being very rigorous in its derivations and systematic approach, options sensitivities leave a wide gap to be filled by heuristics and relatively arbitrary techniques when it comes to two areas: 1, the future volatility, and 2, the right hedging approach due to the former. My approach is "book-based" in terms of an options book, estimating the overall exposure. At any given point, I have positions in multiple expiry and strikes. However, on a minimal number of underlying (10 - 12), I aim to have most of my sensitivities "netted" out and print the critical levels for the positions where I intend to keep the position until expiry and delta hedge those positions. For example, I can take a tesla position of 25 calls with strike 1000 that I am net short expiring on December 17th. The position was entered at the beginning of this year and was initially part of a multi-leg combo on the put and the call side. Most of the positions in the combo have been deconstructed and what remains is the 25 shorted call contracts that I hedge on a block of 125 shares per pre-defined price level with the algo... below is a simple GUI for my algo that is on my desktop for simple and quick actions, initial order carpet and limits and offsets are uploaded via a simple CSV and once started the logic is on the server-side that takes over and handles the positions. It's capable of hedging on multi markets and multi-currency for Tesla, Apple, FB, and Netflix; after the US markets close, the after-hours included the algo switches to XETRA to hedge the position there. I do the design and formalization of my algos and the models but have 2 Dev resources that do the coding. For the Index options, I have another approach involving hedging with options. So not really sure if this is the optimal approach or not, but it has saved me tonnes by not having to hedge more frequently. Section 1: This section lists status updates of the program. After every action taken by the program, a notification is displayed here for a quick view. Section 2: The inputs are displayed in this section when they are loaded from the program. Section 3: The real-time order status of carpet order, as well as subsequent orders, is shown here. Please refer to the strategy for the logic. You can view the order status of BUY LONG orders on the top half and that of SELL SHORT orders on the bottom half of the section. Section 4: The real-time order status of stop-loss orders is shown here. Order types and the offsets are handled by the strategy logic based on market events.
In this specific instance, I exited the entire position, however, there are other ways to squeeze more from the position, but since my intention was never to put any time towards the position post-earnings I exited.
@W-M-A, thank you! I often hesitate to start a new thread because I am not sure if it contributes anything of value, but then someone like you starts commenting and it makes it all worthwhile!
Hm, maybe my trade is not as bad/dumb as I thought lol (or maybe it’s @destriero trade, either way): https://www.tradersinsight.news/tra...ns-market-telling-us-about-facebook-earnings/
We are testing an Earnings Trading Signals system. It had FB for a calendar: 12-Nov'21 325 vs 14-Apr'22 330. Below are the backtested results in sample and out of sample. The long calendars have tested the best especially out of sample. Still a young test.
The results are above. Looking at ~6500 announcements Jan 2020 - July 2021 we optimized signals so they were showing a signal about 2% of the time. That produced an in sample average return for the buy straddle for example of 0.52% in sample and -0.20% July 2021 to yesterday (returns = mid straddle 14-minutes before close before announcement to the next day 14-minutes before the close exiting the straddle at mid profit divided by opening stock price). Selling a straddle yielded 2.01% and 0.19% in sample. The calendar returned 1.31% in sample and 0.97% out. The short calendar had no signals since July 2021 and 1.63% in sample. This test is a proof of concept where we wanted to see if there was any relationship between the data we prepare for the earnings move had any signalling success in simulated trading. Here are the factors we tested: short leg forecast ratio (ORATS forecasted theoretical value / mid price of ATM straddle) long leg forecast ratio implied earnings move divided by historical average move in the stock (near ATM straddle with the residual price removed divided by the average of the last 12 actual moves after announcement) implied volatility divided by historical volatility implied volatility divided by implied volatility with earnings effect taken out If the test looks promising, which they do, we will increase the scope back a few years and use more realistic trading assumptions for commissions and slippage. https://blog.orats.com/earnings-options-strategies-backtest
I will give you my two cents. Don’t even bother testing, your data will be predictive. Why? Because I see more people using this kind of data as it becomes available to masses. So what we have is another self-fulling prophecy: people examine what happened before, make their bets based on the data, the bets create surfaces/smiles (or whatever you guys call it), vol traders make more bets based on this creating pining effect around those boundaries, results confirm. Sure you will have a few outliers, but statistics will show edge. Thoughts?
This is why out of sample testing and actual trading are useful. This is why curated non-obvious data sets are important. This is how edge is developed and determined. What makes the data curated non-obvious? Earnings move estimates use residual straddle valuations based on distributions developed with years of actual earnings moves and IV crush related data. Ex earnings IV modelling utilizes term structure shapes developed from years of data. Forecasted theoretical values utilzes forecasts of HV, long-term cross sectional IV analysis, slope and derivative short and long term forecasts, and earnings moves. My working theory is that the calendar performs better because it takes advantage of the (curated non-obvious) data better than the other strategies tested. The calendar benefits from estimates of earnings moves, where IV will fall after earnings, good HV calculations, term skew modelling, and theoretical values to identify the trades. Testing with superior data is the way to verify, tweak, and ultimately develop a successful trading strategy.