Let’s talk about the NOPE
So, these past two weeks are likely an abnormally good period for the NOPE, and our Google Analytics clearly show it. If you’re not familiar with what the NOPE is, I discuss it more in depth here.
I want to start out by clarifying a few things, because nothing could exemplify the 2020s more than misinformation spreading unchecked.
Let’s start with a bit about myself. I am not a finance person; in fact, by my own admission, I started trading when the zeitgeist got hot around late April 2020. Similarly, I rarely consider myself a trader — I’m very risk averse by nature, and most of my trades capitalize on predicting the market, rather than properly following momentum or buying Tesla or using rocket emojis.
That said, I’m not completely green — my background before my Ph.D. studies (which again, despite popular belief, is in bioinformatics, not economics or finance or math) was in software engineering, and I had the privilege of working in FAANGM for about four years before embarking on my studies.
Why is this important?
I started investigating the NOPE around August, when our first question/use case for the tool (and formula) was not to predict SPY, but a hypothesis that post-binary catalyst events known in advance (earnings being the top example), immediate price movement could be predicted by the delta distribution of the ticker’s option chains. This is potentially implied by delta-gamma hedging as well — there exists, in the case of earnings, a temporary liquidity hole (price decoherence) which causes large volatility spikes right around the binary catalyst event, and at least theoretically there exists a more painful direction for the price to move in due to hedging constraints regardless of earnings news.
Our original data source was Omnieq, which had about the historical options quality data you can expect for a service charging $40/mo (I still have no idea how their Greeks are calculated, and their support/the guy running it never got back to us). I was able from the Omnieq data to backtest it, and woah — it seemed to predict earnings direction on SPY500 tickers roughly 70% of the time!
We learned around late October that well, oops, in some interesting cases our intraday NOPE data (provided by a broker API which I can’t disclose since violating ToS) didn’t agree at all with the deltas provided to us by Omnieq. We ended up taking down the earnings prediction model entirely after some well, very public misses (INTC, CMG — my favorite ones) while we investigated exactly W.T.F. went on.
We then signed up to our current historical data provider, ORATS, and — unsurprisingly — with their higher quality data we saw… no correlation (shocker, I know). More weirdly though, we did verify that Omnieq wasn’t leaking data from the future through various experiments we did during the month of November, so I’m still unsure whether our results were spurious or if Omnieq is doing something completely unique to predict the future. Still would love to know what that was.
Why are you telling me this?
We actually only started investigating SPY as a use case for the NOPE on a lark, around late August 2020. You might remember this time less fondly than I do.
I was vehemently opposed to using it on SPY (it only was for binary catalysts, obviously… right), but we all got a kick out of observing the values shift on SPY. And then we saw something weird, starting around August 28th (back in the bad old days, I first calculated NOPE by hand, then we had a flat HTML webpage serving values — good times). SPY’s end of day NOPE seemed to go up. And up. And up. And this reached a crescendo on that weird day, September 2nd.
We had previously observed $AAPL’s parabolic movement in August, and from my understanding of the NOPE hypothesis, the whole market suddenly looked extremely vulnerable to perturbation. And then September 3rd, we all remember.
That being said, this post isn’t to brag about my success in predicting September 3rd (I do that well enough on my own twitter). I want to come clean on it, especially for the new people who encountered it over the past two hot weeks.
NOPE is still incredibly new, and not very well tested (on the intraday). While I have reams of data from EOD (rather, 3:45 PM, per the data ORATS gives me) which I use to compose my EOD Forecasts, on the intraday as of yet there is no success metric or optimal threshold identified for reversions. In fact, it’s quite possible the success of the reversions the past two weeks is an artifact of a fairly quiet period in SPY’s recent history, where simple mean reversion in price might explain most of the wins (or concordance with RSI, or OPEX madness, or maybe I need to go consult my local astrologer).
I’m thrilled personally to have started this project, and along the way met some incredible people. I’ve always approached it with a research lens — at the end of the day, my question starting this wasn’t “how do I make money?” but “can I, using options data, predict crashes/corrections?” It’s an incredibly stupid question, as I later figured out, especially for someone who barely knew the Greeks when I started collecting data.
I’m super proud of the community we’re building, and the incredible tool individuals like Sean have built to make the model user-friendly.
I’m also really, really happy to see people profiting off the calls, and maybe learning a bit in the process.
That said, please be careful using it for plays (or when I make reversion/end-of-day forecasts on Twitter). I still do not have enough data analyzed to confirm or deny its efficacy intraday, and plan to re-analyze end-of-day data in order to ensure my conclusions are robust (hello, jack-knife sampling on extreme values). It will take some time — I’m still only one person, and I do have real life to attend to. On that note — I am thrilled to have such awesome collaborators, who have done a massive amount of work and helped the project progress to even this point (I am not joking — it would not be here without them).
Finally…
I want to add the ethos of this project and all those aboard is not to make money off this (although, in the future depending on level of involvement, Sean or others may add premium features to the site for profit). I do not and will not personally own the site unless I need to (e.g. a mirror cannot be established without me). First and foremost, I am not here to sell anything. In general day trading communities tend to be predatory to the poor and unfortunately the gullible, and although there are some tools and services out there I do think are awesome, most people are rightfully wary (especially for services which give out plays in exchange for subscription, ahem).
I will work hard to make sure whatever donations collected are transparent and used to support the site, or enhancements as justified (financial data is incredibly expensive, unfortunately). While I may eventually create a trading bot for this if I prove it actionable, in the worst case the formula and research findings have been backed up, and no one (including myself) has permission to take the model away from the community (I am also happy to assist as time permits for individuals who want to implement it themselves!).
We will, however, use the platform as a means to collect money to help the less fortunate. Thanks for the brilliant work of Sean, we’ve started work on NOPE Beta, which in his own words has the following:
- New features/indicators are being built in https://onlyfurus.com/nope-beta
- NOPE beta can only be accessed by unique access key, which will be generated automatically for anyone whom donates towards the selected charity of the month (I picked Feeding America for now)
- I’ve stripped out extra features (stock volume, all indicators, etc) in regular NOPE to encourage people to donate to charity for premium access — NOPE beta has cooler features and I’ll continue to add more, but some of the data is realtime and thus I’m trying to restrict traffic for rate limiting purposes.
Cheers everyone, and happy trading.
