Interesting concept. But won't you end up chasing trades because of the lagging effect between interpreting paper trading results and the systems turning themselves on? You're assuming that there will be follow through and the performance of the live system will mirror the results of the paper trading when you may miss the good trades entirely. For example, if the paper trading showed that a trend following system worked well on Tuesday 10/28 (when the Dow went up 900 points) and this system went live on 10/29 (when theoretically the market is going sideways) then you'll get a different result.
I agree with that. I currently believe that any kind of "predictive approach" (after all you are kind of assuming a sort of "inertial effect", so that favourable conditions should continue for a while) is not usually rewarding, or better than random trading. I am more for systems which, always based on market conditions, try to obtain a profit ,say, through a more "mechanical" approach, where you do not "run after" prices, but wait for them and create the conditions to maximize the possibility to obtain a profit by appropriately closing and opening positions in the right directions. I like to represent this, as i have described on a web page, as a "probabilistic cage" or trap", built (and estimated) in such a way that it's rather unlikely that a price curve can "escape" leaving you sistematically with a loss. T
I fully get where you are coming from. It is better to identify a 90% probability of an event occurring and placing a bet on it with good money management than waiting for it to occur and trying to chase it. I like your term 'probability trap' -- that is exactly what it is. Just gotta watch out for those fat tails when we are wrong...
Corey, Here's another way to think about what's been discussed here: Consider that you've defined the problem correctly and that your proposed solution, defining a market condition isn't that far off. Then what's left but the proper analytical tools? It's unlikely that any of the appropriate analytical tools are available off the shelf. For example, it sounds like tomasso isn't using technical analysis [from his website]. For instance, regarding appropriate tools, consider the limitations of using standard deviation as a measure of "volatility" for a trend-following strategy: 1) std dev is a measure of variability from the mean - is that the kind of vol you care about?, 2) the data series you're evaluating is likely non-stationary. Just sthg to consider when looking at your tool kit. -BlueLou
This an interesting remark, as the kind of volatility we need for successful trading is not that necessarily represented by the variance, as a large variance can be achieved for instance both with very irregular price curves or with smoother ones. I think we are more interested in some "specialized" form of volatility, which is difficult to express in an closed form formula. Perhaps a set of indicators which take into account the "kind" of volatility can be useful. And the idea by Corey of classifying the market, perhaps can be expressed as an attempt to classify the current "kind" volatility. Another idea i had recently, and that sometimes I would like to implement in a project, is that of creating a sort of "normalized price representation" (by appropriately shrink or magnify dinamically the original prices) which would allow to trade by using an universal set of parameters, clearly much simpler to be estimated and maintained. In this way the volatility concept could be somehow incorporated in the data through a form of normalization, and the trading model could not care about it. Further, this would allow for a further "metaparameter" which could for instance shrink or enlarge easily the relative positions of the trading parameters, to obtain different trading styles: more conservative, more risky, and so on ... T
Tomasso, It wasn't rigorous but I tried doing something like this where there would be a regime switch to a different tick rate depending on the information content and irregularity in the data series. I was looking at FX futures at intraday and higher frequencies. Here's what I concluded: As the frequency (tick rate) increased, I found greater regularity given the amt of information; but, economically it wouldn't have made sense to pursue this w/1 guy and a PC infrastructure. Maybe it was bias in my statistics? In retrospect, I think I started at too high of a resolution. Where I do plan to give this another shot soon is in determining the optimal scale for an individual instrument and measuring how this evolves overtime. At this point my hunch is that the optimal tick rate evolves somewhat slowly, at least for my strategies, maybe an AR(5) or AR(10) process based on the total number of ticks in a daily bar. Have you done any work on optimal tick rates? Sorry, I think I'm getting too far off the topic. -Lou
PCs nowadays have no problem doing that kind of job and you can do a lot of processing between a tick and the following. Yes I found that tick rate is something that any day trading system needs somehow to take into account. Although it is not mentioned much around. Isn't it? [Clearly, I am talking about day trading.] The highest avg tick rate is always from SKF, usually followed by AAPL. Perhaps it is just by chance that skf is usually the best sec for day trading (?), but in any case trying to day trade anything under 15-20 ticks per minute is probably just a suicide. Isn't it? I have "practical" experience, but I must confess my ignorance about the mechanism which is really behind the tick rate. Does anybody know if there is also some correlation between tick rate and average volumes ? [Also day ats should always keep an eye on the spread.] T
Tomasso, I think Engle's papers on autoregressive condition duration (ACD) discuss tick rate vs. volume relative to price changes. I recall that he concluded that transaction intensity (tick volume) is much more closely related to the duration b/t price changes than volume is. If you're familiar w/GARCH you'll pick up on the ACD stuff quickly. Believe it or not, my algorithms really stress the processor and they eat up memory b/c of the amt of historical data needed. In fact, I can't do anything strategy-wise intra-tickbar b/c of the calcuation intensity. I've just started to look at equities in the past week or two. So far, I've been trading only futures but the intraday vol in the financials has exploded and some of the spreads have widened so I'm looking to expand to equities now. Are you just looking for equities w/the highest tick rate - as a proxy for liquidity - or are you finding other info based on the tick rate? -Lou
Engle and Russel seem to be talking (in "the analysis of high frequency financia data" paper) about transaction rate, although they also talk about ticks. I think this might be confusing. [I have also written some academical paper, but one must keep in mind that these kind of papers are meant usually for career purpose. It's quite difference when your model (and bug in code) cost you actual money. If statisticians should pay their own money for any formalization or distributional assumption which only works in the world of ideas and not in the real world, there would be very few scientific journal around, and very thin! ) ] Frankly, as to the tick rate, the main use I have been doing of it, is to exclude dangerous trading situations, and not to determine opportunities. It's clear that day trading needs data, and a low tick rate does not help at all. As to the question, that you probably have in mind, whether some indicator based on acceleration or deceleration of tick rate may increase the strategy performance, actually I do not know. The answer is probably no, but it's a question that could be investigated, just in case. Anyone has some information on that? T
tomasso, I understand now that you're looking at the number of ticks per unit of time, "tick rate", for liquidity purposes. Transaction rate is the arrival time b/t ticks. ACD is meant as a tool to analyze whether changes in transaction rate contain information. The evidence suggests that they do. I disagree w/your comments on academic research. If you know what you're looking for you can separate the good from the bad research in a systematic way. -Lou