Today while the market was dropping and I was losing on many positions (partially hedged via SPX, UVXY, VIX, and puts on underlyings), it's also a good day to review which options positions/combos managed to self-hedge and stay in profit. Found some bullish option combos I'm holding that I really like, as certain bullish positions managed to turn profit today even as the market was tanking. Here is a bit complex self-hedging bullish options combo on FedEx. It's basically a long/short strategy on a single stock, removing the need to go long on some stocks while being short on others. It also makes me suspect that retail traders can have a small advantage over market makers, as we can choose which options to trade, while MMs trade pretty much everything and spread their risk more evenly via delta hedging. Now the goal is to zoom in on similar types of option strategies.
Interesting. Question - how would the above complex-options position chart look if FDX had risen slightly in the last two days, instead of falling? Would the strategy still be making money? After seeing your post, I came up with this in a few minutes..... similar self-hedging on the downside, but if FDX had risen, my trade would be flat, unless the rise was significant.
Sure, you could do this even with basic delta hedged combos (possibly with small delta towards bullish or bearish direction) or a strangle. My trade took some careful work, analysis and testing, and should be profitable on the way up, and that’s how I made most money last July-August. On bullish side the profit grows either with large move up or with slow move up over time. The biggest risk being no movement or very slow move down, or strong volatility drop that would decay my hedge. Generally this is a bet on the price moving either direction at some point before January, preferably bullishly. But there is no way to guarantee results in either direction, as volatility and time also play a role here, so I’m just observing such setups, writing down notes, and thinking of ways to trade them reliably and with decent profit factor, accounting for some losses. I’m also tinkering with backtesting these, but still have a lot of work on that, for now just getting basic and overfit confirmation that this may be generally profitable/promising setup.
Cracking the smile. One of the main pieces of the puzzle for options backtesting is properly estimating theoretical option prices, especially DOTM, including accounting for IV smile. So this is the "math" I've spent most of the time on recently, including testing solutions from various research papers. I wasn't happy with any of them as I couldn't get estimated option prices anywhere close to what they're actually are. But I have a theory that the IV smile for DOTM options does not really come from "pure" IV but most of it comes from necessary impact of arbitrage-free option pricing adjustments. So for example when someone buys thousands of SPX puts then the IV may change for that strike, but then all the models will not only recalculate IV and pricing for remaining options, but also adjust those calculations further to assure they're arbitrage-free. This, in my opinion, creates more impact on the other OTM option pricing than just pure demand, also accounting for majority of the "smile" effect. So based on this assumption I implemented option pricing calculations that are mostly driven by non-arbitrage requirements, with some adjustments/recalibration using available option prices/strikes. This allows me to take OTM option prices of any 3-5 OTM options, and calculate all the further OTM option prices, without Black/Scholes, without IV, without Greeks, and without approximating anything. Just purely analytical/logicical solution that tells me what the DOTM prices should be, based on sampling of a few nearby/available option prices. I'm not sure whether my approach is common or unique, but at least I think I now have near-optimal pricing formula for DOTM options, which not only calculates what the option prices may be, but what they should be. Still lots of work with all this, but I feel like I'm slowly getting through pieces of the puzzle, one at a time. Couple examples: a) SPX DOTM Put price forceast/estimate based on sampling of Put prices for strikes 1800-1900: b) IBM DOTM Put price forecast/estimate based on sampling of Put prices for strikes 100-110:
Since modern vol models calibrate to prices almost arbitrarily well, you've either coded the papers up incorrectly, or applied the calibration improperly. Or maybe you've somehow chosen the rare papers in this space that get it wrong. Code for, e.g. stoch vol, quadratic normal vol, rough vol.... models, is widely available on github and elsewhere. Try re-running your calibration using other's code. Alternatively, post an SPX option chain (as csv file) you've had trouble fitting and I'll have a look at it. Model-free parametric fits of price curves/surfaces based on no-arbitrage constraints? Common, far from unique. There are advantages to coming in to the options markets as an intelligent, talented outsider with no industry experience or relevant academic knowledge, but there are downsides too. In this case you should not be reading academic research or even books, but rather the notes put out by JPM, Barclays, Santander, GS, etc.... Either that or partner up with someone who knows the space (not Aquarians!).
Thanks Kevin. And you're right, I have tendencies to look at exotic solutions/papers that no one else is looking at, while omitting stuff that's actually used in the industry. Partly due to not knowing what is being commonly used, partly due to not understanding some advanced stuff, partly due to papers themselves not presenting intuitive solutions (just ways to optimize and curve fit data). Though I'm also concerned with being complacent and using solutions/functions that I don't have to understand internally, and having tunnel vision and not seeing potential edges. I understand that option prices can be curve-fit or derived without BS or IV, though I still suspect I may have somewhat unique approach because I haven't found any literature about it, while my approach doesn't use optimization, coefficients, or curve-fitting directly, but is more intuitive and explainable than what I've seen elsewhere. Though I understand that in the end this is still curve fitting, maybe indirectly. I'm attaching sample SPX Puts I've used, snapshot from a week ago, for expiry Jan-15 2021. CSV columns: strike, price, IV (not used, just in case). I used these in my above example to forecast DOTM put prices below strike 1800, using only five options/prices of 1800-1900 strikes. Not using Calls yet for call/put parity, and no other expiry dates for now.
Didn't have much time to play with the option smile/pricing last week, just went back to using quantlib with Heston model calibration and some custom calibration points, but I have to run more tests before backtesting. My main problem may actually be in just totally invalid pricing on some less liquid options, as well as UVXY, and nonsensical bid/asks on some OTM options on larger stocks like SQ. And not sure yet how to approach VIX with BS, but that's where I make most money recently, using my own model. My previous SPX hedge lost all its value due to SPX recovery, so that's gone, though I've managed to sell 25 of them to extract a few $hundred profit. I decided to check my account balance (after trying not to, to focus on each trade) and I'm up 5% over last 30 days. But too much of it still due to luck, while my risk and drawdowns are too high (also 5% drawdown in the last 30 days). Part of that due to testing variety of ideas, and setting up trades as a way of testing and research. My system now spits out 100K-1M option trades ideas at any given time, which aren't possible to trade with limited budget, so I have to spend hours reviewing, selecting, and testing/trading some that seem best using various measures. Checking where I made most money since September, it looks like from VIX backwardation arbitrage (about $18K), where I actually do have more advantage than luck. Since/if VIX is already mean-reverting, then mean-reversion of VIX futures backwardation is pretty much a sure thing. It's just me who is never sure whether my model is fully right, so I slowly scale into those trades, sometimes missing best opportunities. For the recent December/February backwardation I was trying to slowly scale into trades, but they started running away from me and I'm up only $1K: I'm also doing well with UVXY options while able to limit risk, but also didn't put much size on to keep that risk small. Where I make most theoretical progress recently may be in my system coming up with complex option combos/trades that look amazing in some aspects, but I'm not sure yet how to evaluate the risk. May just need to trade them. Here is the profile of one that I may describe as "time arbitrage" (though there may be a better name for this), as it bets on the fact that Brownian motion is a motion and doesn't stay still, therefore it covers variety of price and volatility movements over time. If SPX goes up then it profits soon, if it goes down then I may either wait, or profit from increase in volatility. Drop in volatility doesn't affect much of the upside either, while it can wait up to 2 years for the price and/or volatility to move either way. There is also room for further adjustements and hedging.
I needed something that helps me decide when to make bullish bets, so I went over almost 1 billion stock trading strategies I've gathered over last few years and picked a few with 85%-90% accuracy (win rate), avg win larger than avg loss, never having a losing year and barely even having any losing months - since 2007. Are they overfit? Hell yeah, like the most overfit strategies you could think of But the overfitting, in this case, comes mainly from inability to tell how much to buy and when to sell or stop, as those are additional unpredictable variables on top of already difficult to predict market direction. However, these strategies do seem to logically detect certain level of bullish market sentiment, and therefore may have a degree of predictable "power". We'll see, while I'm going to use them to make some option trades myself. Also starting to Tweet them as reference trades for anyone else wanting to follow and see when the market conditions are ripe for bullish moves... Here are some initial trades my (new) bot tweeted last night: Ongoing stats /list of trades will be published here (free): https://deustrader.com/zeus-longus-1.html I'm also setting up separate Twitter account for these: https://twitter.com/SentimentMarket
So it may all come down to continuous casino-style betting on stocks and options, using Kelly-sized bets. There really isn't anything else that math gurus at Rentech could or should be doing. You’d make specific bets using options, usually without strong direction but not necessarily directionless, then keep continually re-betting, which may be done by scalping shares against held options and odds, vs hedging to zero delta. Although I guess delta-hedging, in theory, is also based on odds and since it makes money then someone could say they’re already doing it. The difference may be in being able to increase the odds by making specific bets and with proper size, vs MMs simply taking all bets and having tiny advantage. This approach can also be used with stocks, though it may simply mean holding Kelly-sized shares/positions, occasionally readjusting. This should work somewhat differently from modern portfolio, and can even work on a single stock. Now it’s just the matter of estimating correct odds, and I think I’ve got this, which also gives me the end goal and vision. I’m starting from beginning and redoing everything I’ve done in the past few years. First I’m building a new stock strategy extraction/analysis/testing system for stocks because it’s simpler than options and I can test and fine-tune it before using it for options. For now it's basic statistical analysis that allows calculating Kelly bet size for holding a stock, but it also allows me to extract certain natural behavior of stocks (Markov?) and produce theories that later can be tested and validated. This also results in zero overfitting, to the point that now none of my newly extracted stock strategies can beat the basic strategy of simply holding shares... So in the end stock picking and holding shares seems to be a much better approach than trading, though not necessarily. In one way I now see clearly that holding shares means that you’re always on the train and don’t miss getting back on it if you exited earlier. While trading means getting out of Amazon or Tesla, missing a move up, then never catching them again at the previous/lower price. This is why trading doesn’t work for retail traders, especially as they just want to get their favorite stocks "cheaper" and keep missing upward moves. In the end successful trading means always looking for a greener grass, and that’s what makes it so difficult. So now I’m ending up not finding too many stock trade ideas and strategies with my new system, except for just picking stocks that naturally perform well, such as SPY. But this is a great starting point and a "clean slate" because now I can finally see true problems and focus on solving them, vs previously most problems could be solved by "fitting" a system to any data. On another hand, I'm also starting to see some potential trading/betting opportunities, sometimes in surprising places like trading UVXY long. It may not be worth trading due to low RoMad and low median gain lower than avg gain, but at least I'm getting new ideas for hedging, which potentially can be enhanced later using options. I'm also now seeing and learning why and how high leverage can be useful, especially when able to get high RoMad but low avg gain, though I don't think I'd want to trade that way. Anyway, now I see there aren't too many sources of alpha (without external information) in the general universe of stocks, except for general odds of holding some, or even holding small slices of majority of stocks. However, now I have a base system that naturally prevents overfitting and makes it very difficult to make money trading, which is a great starting point providing actual problems to solve, vs my previous stock trading system where anything could work with a bit of fitting. And when applied to options, the new system actually gives me a ton of stuff to work with. Now my strategy development consists of extracting natural behavior out of certain stocks and options, as well as potential trading methods & signals that naturally fit the behavior of those stocks (hidden Markov?), which then become theories that can then be tested and confirmed. And I even started using RealTest to validate a potential strategy, and see it as quite a nice, useful tool. Though even without overfitting, the behavior of stocks and markets may always change, so having temporary advantage isn’t the holy grail, at least for stocks. Trading-wise, I’m slowly making money trading options but not reporting much recently because I realized I’ve had a false start earlier, and still a lot of work to do. Basically I initially started by testing my new options strategies and making some money and being able to defend my positions when the market wasn’t moving much, then suddenly the market started going up and I was quickly making money and putting new trades on every day, without realizing how much it’s due to the market vs my strategies. Now I’m back to slowly grinding forward and let the market work as slowly or as quickly as it naturally does. I’m up minimum 5% in the last 30 days, seeing 10% at one point, but it fluctuates a lot due to still being sensitive to market moves, and due to holding many OTM and LEAP combos that show different value each day, for example LEAP OTM butterflies showing value between $1.50 and $8 depending on the day, while I can’t buy them cheaper or sell for more than around $5/ea. The only problem is that I’m spending too much time... trading, which means staring at various option ideas, analyzing them, picking some to trade, then analyzing stats for them, defending losers, deciding when to close positions, etc. This all would be great if I had interest in trading, but I'm having more fun figuring out solutions to problems, and just exploring new territory. At least after trading options based on probabilities for a while, it’s become obvious that the only way to trade on large scale should be mathematical (though not necessarily using the same math that others use), and the best way to do that with options may be by continuous betting. Once anyone knows the odds on their various setups and trades, then it all comes down to using Kelly’s criterion for proper position sizing. Though right now I have too many setups to choose from, so that I don’t need to bet big on any of them, just taking small option positions in various companies, ETFs and indexes. At the same time I’m starting to pick a few simple short-term option strategies that I like for their liquidity, repeatability, scalability, and great potential overall for trading.