I agree with AAA. Not backtesting, but more data mining for statistical correlations although that doesn't sound as sexy.
A good point. The statistics we provide give you a sense for the distribution of historical outcomes when some indicator value was more extreme than it is today... meaning we give you a sense of the historic worst case scenario, best case scenario, etc. The idea being that you have a sense for what to expect when using an indicator. As you point out, any events that are not captured by historical precedent can't really be quantified in this way, and that is a drawback with using systematic signals and historical analysis in general.
I guess I'm having a hard time wrapping my head around what you're saying. There are services like Quantifiable Edges that do something similar to this, but he does it in a way that makes sense to traders. I can't see what value you are adding by listing a group of correlations and outcomes. Again, I don't want to be unfair and I apologize if I am misstating your product. You obviously have access to some serious computing power and data, and I'm not saying there is no value there. It' s just that I don't understand how I would use it. Traders would typically start with the idea that we want to isolate a set of conditions that historically has lead to profitable results. Maybe i don't understand what you're doing, but you seem to be going at it the opposite way. You list a group of conditions and say these are the results, and those results are all over the map. Even if the outcome is statistically significant and positive, it would be hard to trade without knowing how the results were obtained, ie what was the drawdown, how were the trades distributed, did the trades generate alpha or were they just market-drive? I'm alway sinterested in projects like this and will continue to follow it.
AAAintheBeltway, Really appreciate your thoughtful comments. I think I'm doing a bad job explaining the product. If you'll bear with me, I'd like to take another stab and then get your thoughts on whether the value is existent/clear. What we do: 1. For a single stock, we track what the values of a bunch of indicators (right now RSI, returns, highs/lows, Bollinger bands). Lets take the example of the ticker GORO. We produce a table like this everyday for GORO (for now please overlook the fact that is says "backtested") 2. What the table shows is what current values of each indicator are for GORO (first column). To the right, it shows what tends to happen to GORO stock historically after the indicator value for GORO is more extreme than it is today. For e.g the first row says that the 5 day RSI for GORO is now 16... what the rest of the row is saying is that the RSI (5 day) for GORO is less than 16, 4.6% of the time, which represents 78 observations and that for the subsequent 10 days, GORO on average tends to return 4.1%, with a best case 10 day return of +54% and a worst case 10 day return of -15%. Also, GORO tends to return a positive value 72% of the time. 3. So the rest of the table shows similar statistics for other indicators. The question we are answering is "what do a library of indicators currently read on GORO and historically what tends to happen to that stock after the indicator value is more extreme than it currently is?" 4. We produce statistics like the one above for GORO everyday for over a 1500 stocks. Now, here is the key: we allow you to search through that huge database of results every single day in almost any way you want. So you can ask a question like "show me potential trades across all stocks where the RSI for a stock is at an extremely low value and is historically followed by a very positive return?". Thats the table below: So how would traders use this: 1. Idea Generation: It allows you to filter through things going on in the market in a totally different lens than what is currently available to anyone who is not a "quant". Most screeners allow you to filter by indicators.. but NOT by the result of what tends to happen to that stock after the indicator is at its current value 2. A quick way to see what a library of indicators says about a stock and whether a particular indicator has historically worked for that particular stock Would love to hear back from you about what you think of this. Really appreciate your thoughts! Thanks!
I appreciate the explanation, and the product is interesting. Here's the problem I'm having. In the past I have done a lot of backtesting. One of the first things i proved to my satisfaction was that blindly buying on the basis of an oversold indicator reading, eg RSI<20, was not a viable approach. I have to assume that while you generate a list of stocks that buying oversold RSI would have "worked" on in the past, there are many more where it would not have. So what is different between the ones where it worked and the ones where it didn't? How do we know it isn't just some random occurence?
Hi AAAintheBeltway, To put things in context, this product evolved from requests by a group of long time professional active traders who were NOT quants or systematic. These traders looked at things like RSIs and believed based upon anecdotal evidence that they added incremental value to their process. They were very excited to use this product that we put together for them to screen for opportunities, and so we wanted to test whether there was a larger market for it. Hence our asking for input on this msg board (we're not trying to sell anything yet). Now, you bring up a key point. If we put an RSI indicator in our product are we saying "we think RSIs work in general"? To be clear, we are not comfortable making that statement. However, we noticed that products like Bloomberg allow you to plot RSI and backtest based on it, and so there is a question of whether they are also effectively saying "RSI's work in general" or are simply providing things their customers have asked for.. You could argue about this both ways... but this is one of the things we wanted to test and this is why your feedback has been very helpful. We do believe there are ways to use the structure we have created to create something valuable for someone like you. We're interested in using the same structure I outlined in my previous post to simulate published academic work that is inaccessible to anyone not willing to work their way through the math and read/understand/simulate that work. Do you think that is something that would be useful to you?
I doubt the system takes into account holidays which tend to shift from date to date unlike the regular calendar.Pesach,for e.g.,etc...
I note that my previous post was censored - the animated ROFL emoticon was removed! Never mind. To be clear I don't believe your product has any value because any prediction method will fail, and also the use of indicators like RSI, stochs, macd have no value - they simply don't work. I say that as a result of a study I did a few years ago using share price data from yahoo finance for numerous stocks, using my own programs to find profitable combinations of indicators. The programs I created used good old fashioned fortran for speed and genetic evolution to search for any combinations of profitable indicator parameters. I flogged the indicator methods to death and never found any consistently profitable swing trading strategy using them. Also realise as I do that anyone with a little programming skill can get free end of day share price data from yahoo, google, quotemedia etc and do their own scanning using their own specific criteria, the scanning process just takes a few minutes.
Hi Mysteron Thanks for your thoughts. Much appreciated. [, QUOTE]Quote from Mysteron: I note that my previous post was censored - the animated ROFL emoticon was removed! Never mind. To be clear I don't believe your product has any value because any prediction method will fail, and also the use of indicators like RSI, stochs, macd have no value - they simply don't work. I say that as a result of a study I did a few years ago using share price data from yahoo finance for numerous stocks, using my own programs to find profitable combinations of indicators. The programs I created used good old fashioned fortran for speed and genetic evolution to search for any combinations of profitable indicator parameters. I flogged the indicator methods to death and never found any consistently profitable swing trading strategy using them. Also realise as I do that anyone with a little programming skill can get free end of day share price data from yahoo, google, quotemedia etc and do their own scanning using their own specific criteria, the scanning process just takes a few minutes. [/QUOTE]