There are many details that are unclear. Are you back testing option trading strategies that are written by a "USER" (near boundless in form and complexity), or are you limiting to specific canned strategies? Is your backtesting done with only EOD option data, or finer granularity (Trade Data, 1 min, 5 min, etc) Are you interested in collaborating on development challenges, such as algo development? Pardon all the questions. My personal interests is in developing something I can use for my own research, and trading, not specifically for developing a product. However, there is likely a bit of overlap. My guess is due to the amount of effort to pull this off, those that have been implemented previously are in Private use.
>> Are you back testing option trading strategies that are written by a "USER" (near boundless in form and complexity), or are you limiting to specific canned strategies? So far I've tested single options + canned strategies like spread or butterfly. But I support any combination of options, provided you can describe it somehow. >> Is your backtesting done with only EOD option data, or finer granularity I'm using EOD data. >> Are you interested in collaborating on development challenges, such as algo development? Yes, definitely. I'm also interested in pricing model changes, ideally being able to describe that mathematically. But I also have models where I'm just describing stuff computationally. >> My personal interests is in developing something I can use for my own research, and trading, not specifically for developing a product. Mine also, initially, but as the complexity grows, I feel like I have to collaborate with others. >> My guess is due to the amount of effort to pull this off, those that have been implemented previously are in Private use. Yes, also I think so. Maybe, probably, start public and then go back private at some point.
Interesting subject. IMHO, price-level back-testing is kinda misleading for options. You going to arrive at the usual "easy to make money by selling premium" kind of thing. At the very least, you should work out a way to separate underlying dynamics from the option pricing component (i.e. that increases your sample space significantly). It is especially true if you are trading options that have maturity longer then a week - this way, you can simulate many more "trades" and resample. It's also useful to back-test your option strategy using options priced at fair vol (both actual realized and preceding realized). It will tell you how much you are relying on richness of risk premium and how much you gaining from the underlying pricing dynamics. For some things, like S&P 500 or DJI, you can actually go back as far as 1895 and go through a variety of stresses. For portfolio strategies like directional dispersion it will actually tell you where the alpha is and how persistent it is.
SLE, I don't get everything you're saying but this just proves that more people means more expertise. Regarding fair vol: I'm not sure what you refer to when you say "actual / preceding". Like current computation time window vs previous one? Directional dispersion: never heard of it until now. It's one of the countless possibilities, sure.
-> I'm using EOD data.<-- I suspect this will prove problematic. I think it wise to ask yourself the question "How accurate is the Back Test?" Also then consider the "edge or profitability" of a "backtested" option trading strategy. This "accuracy" must be adequate to handle the precision necessary! Note: I would expect, most people, including myself, who have high interest in option trading backtesting are primarily sellers of premium, and not merely directional traders, so the margin of error in the backtesting must be very small. By limiting early work to an index, such as SPX, it may be practical to purchase the data to reduce the error to your equations for slippage (which can be easily controlled). I have done some preliminary work in creating an IV surface map from available data, which is then used to apply to option pricing models to approximate the real prices. This is not easy, and the complexity may greater than the ability to merely purchase the real historic data. The more advanced option trading strategies are likely greatly influenced by inflections in the Skew of IV, which will be difficult, if not impossible to model precisely. You may prefer to be able to quantify the skew, and allow the trading algo to make use of that for timing or picking strikes for adjustments. Note: I'm still trying to get my head around this so not implying it is impossible, but merely lower hanging fruit seems available if goal is merely to backtest with accuracy. (lower hanging fruit being Live Vol historic data) I'm curious for your response! (BTW: I agree with your other responses)
@stepandfetchit: >> I would expect, most people, including myself, who have high interest in option trading backtesting are primarily sellers of premium, and not merely directional traders I don't do directional at all at the moment, although I have plans for trying that some time. But right now I just don't have the time to put down the mathematics and then to code it. >> By limiting early work to an index, such as SPX I started with a single index indeed, but now I'm covering several of them and also stocks. >> reduce the error to your equations for slippage The margins I'm using allow me to ignore that (I hope ) >> I have done some preliminary work in creating an IV surface map from available data, which is then used to apply to option pricing models to approximate the real prices There are lots of ways of doing that and as far as I see it, here's the key between making money or not. >> than the ability to merely purchase the real historic data Well, I did purchase EOD data. I started initially with realtime data but found the models get too complex so I moved back to just EOD. >> The more advanced option trading strategies are likely greatly influenced by inflections in the Skew of IV, which will be difficult, if not impossible to model precisely. I haven't found so far a perfectly good explanation for the skew, models like Heston or Merton are possible candidates but after spending a lot of time with them I reverted to simpler / other stuff. Maybe at some point I'll use them again, with modifications. By "more advanced option trading strategies" I understand hedging for higher order greeks, not just delta. For instance delta-gamma hedging does reduce a lot the standard deviation of the realized PNL compared to plain delta-hedging. >> You may prefer to be able to quantify the skew, and allow the trading algo to make use of that for timing or picking strikes for adjustments [...] lower hanging fruit being Live Vol historic data You lost me here I suppose I'd only know what you mean if I get to know your model and strategy in detail.
For back test, I could get data for stock price, risk free rate, dividend rate but not for IV. I couldn't get IV for specific strike prices since my broker only provide one historical IV # and historical actual volatility #. So, I was forced to use a power function for skew based on some guess work. Using the above, I back tested mechanically selling calls and puts on all the stocks I owned and compared with my actual trading results. Most outcome were not as profitable as buy and hold the individual stocks and reinvested the dividends. So, I am now a directional trader. I am trying to expand into trading non directional and volatility... thus need more back testing ideas. Appreciate if you can provide some suggestions. Regards,
IMHO: My advice: Anyone tells you to do x, or y, beware! If it sounds too good to be true or it if is easy, .... beware. If told only good things about a trade, ignore that person forever, as they are exceeding evil or misguided. There is little if any "low hanging fruit". Your brain, is your best friend! Only you will be able to determine what fits your trading/investing style or requirements. (I think we are all different, but that should not be considered a bad thing, but just means the subset of traders with compatible interests are perhaps a smaller group than we would like) Some people like the "Investools" methods, which seems extremely foolish to me. -- I went thru this, and consider it a weak primer, but perhaps needed. Some people follow Sheridan Mentoring, which may be OK if you like a little bit of shooting from the hip. There is some good info here. I now tune in to CapitalDiscussions, which, for me, is a closer fit to my interests. Many here are full time traders, or retired, and trading. Backtesting could possibly be putting the buggy before the horse. This may not be the best way to develop a good trade, but may be a good way to refine/improve a trade. (IMO)
>> stepandfetchit: Anyone tells you to do x, or y, beware! If it sounds too good to be true or it if is easy, .... beware. >> Backtesting [...] may not be the best way to develop a good trade, but may be a good way to refine/improve a trade. Exactly. >> ironchef: I am trying to expand into trading non directional and volatility... thus need more back testing ideas. Appreciate if you can provide some suggestions. For me, it starts with the fundamentals. A model / strategy needs to make sense from a mathematical / common sense point of view. Then it needs to be tested on the market. I've discarded a lot of models, as beautiful as they were and feeling sorry for the effort put into understanding then implementing them. But they didn't work, or at least not on that market. Eventually I came to the conclusion that no model fits all markets and also markets change over time. So backtesting first determines what model(s) currently fit the market. Live trading needs to both figure out when a model no longer fits and to control the risk such that one doesn't lose all the gains in one blow. I'll discuss about backtesting ideas but first I need to clarify a few things.
Two questions: 1) Do you think trading is something where everyone wins or it's a zero sum game? 2) Do you think you can open-source your model/strategy/code and still trade profitable using it afterwards? I think: 1) It's zero sum. And when it's not, competition drives it to zero anyways. 2) No, you can't. On #1, You can look at options as a form of insurance, so you can rightfully say that the relation between the option buyer and the option seller isn't necessarily zero-sum. Compare it to the relation between two poker players: one has to lose in order for the other one to win. But at least in theory, the insurance buyer is winning the protection from risk, and the insurance seller is winning the premium. From this point of view it's both fair game and a mutual win. The problem is that #2 kicks in: even if the insurance is not zero sum, you can't disclose your formula, or competition will drive you out. Even when option sell prices are very close to the zero-profit point, a big insurance company like Goldman Sachs can still make money because it can afford to have the volume which makes up for the losses. A small company just needs higher margins in order to survive. And those higher margins do exist and do result from proprietary models/strategies. The whole profitability in options revolves around knowing where zero is, because everything you add over that is profit. What I have is the knowledge / code who at least gets the zero right. Actually it looks like it's profitable, but I need your feedback on it. Will continue with that, but in the meantime I'm curious what is your position on #1 and #2.