jurotraderslab

It's about the same as what I found out based on my extensive backtests. Scaling out all at once in general is the best you can do in terms of overall profitability but the equity curve may not be as smooth as let's say scaling out for partial profit (or setting stops to break even, etc.). This, however, is only true for 'general' strategies, there are specific situations where it is better to do something else (anyone has a momentum based strategy?). Needless to say, and that is the main point of the article, the psychological aspect of scaling out is attractive and most discretionary traders employ it in some way but I would argue that for an automated system, it's more attractive (simpler programming) to just get out all at once...

I guess I didn't clarify my question properly, I meant significance with respect to price move (not volume traded). For example, consider a sequence of trades that plays out within 30 seconds such that your indicator shows a big spike (> 50000) and during those 30 seconds the price only moved 2 ticks. Now, would it show the same value if the exact same sequence of trades played out over a 10 tick price range? Here I just mean timestamps and trade volumes, not prices at which trades happened, so still a 30 second time interval with identical times/volumes transacted as before but "spread out" over a 10 tick range.

UB, still back to my originally posted question, do you show increased intensity only if it's concentrated in a small area of price? If yes, do you put limits on what small is for various instruments?

Steve, I think there is a nonempty intersection between your and UB's thoughts, however, you don't (want to?) buy into the concept that things can be automated ("seen by a computer" instead of watched by a human...). Believe it or not there are some sufficiently smart people who code what you "see" and then just enjoy watching the computer trade for them. Of course, many things are hard to code but what you mention (tug of war on tape etc.) is certainly something HFT have no problem with capturing.

One question that comes to mind immediately: do you distinguish between spikes that occurred on a fast directional move over an extended price range (let's say > 4 ticks in ES, i.e., > 1 point move to "stop" the price) and spikes that only appeared to play out in a 1-2 tick price range? The first category might just suggest desperate unwinding of positions (or jumping on a move), the second one seems more promising as far as reliability. Does your indicator pick this up? On a related note, I find it surprising that someone would be continuously hitting the bid (in case of intense selling near a future bottom) which is being continuously replenished. How would you explain that behavior?

Could you show some pictures of other markets, say currency futures like 6E, 6B, or rates (ZN, ZB), or soft commodities? Does it work as well as on SP500? I suspect (speculate...) that in equities big institutions buy/sell futures (in an automated fashion...) before they turn on their cash buying/selling activities so it makes sense that these spikes could capture reversals (as the index really only follows the underlying stocks). However, in a currency like 6E the futures market is just a tiny fraction of transacted volume in EURUSD (let's consider futures and cash the same for now...) so one would think it shouldn't work as well. Do you have observations (or better yet hard statistics) that confirm it's universally valid across markets?

What do you mean exactly by "exhaustion" (how does it show in your picture)?

That is great, could you show a picture for an entire day, preferably for a trending day? I would expect a trending day would have many false spikes. This is not to undermine your efforts but I would like to see it that is the case. Also, based on your posts from the past, you normalize the spike indicator to time of day typical activity (time of day normalization, of course, this includes instrument normalization too I expect as every instrument has different typical volume figures). Do you also normalize it dynamically, based on how the day is progressing with respect to longer term volume statistics?

UB, I wasn't arguing that it doesn't work, however, what I was saying, the trade intensity peaks also happen at times when market continues its directional move. It would be fair if you also showed an example of when the indicator peaked but the market didn't turn.One cannot be sure that the current peak is the highest of the day and therefore any peak beyond a certain (albeit potentially dynamically adjusted...) threshold would have to be treated as a trading signal.

Steve, you are right, no one can tell exactly when institutional players decide to step in, the question for UB is whether or not he can take ALL these set ups as shown and trade them automatically based on a fixed set of rules. One trivial example: when both trade velocity and net new trade diverge on a local swing high/low, take a trade with predetermined stop/target. Does this setup (or a "sufficiently smart" modification thereof) produce net positive alpha if it's executed automatically? If yes, then he's in business, but I guess it may not since if it did, he would be already trading it instead of writing about it on TL. I am just writing this because a) I think there is some value to what UB says b) it's not as easy as UB seems to make it (local extremes are only a subset of all situations in which these divergence setups occur, the profitability then depends on how you can separate these setups from the ones which are followed by a continuation as opposed to a reversal).

Thanks Tams. Could you also include an equity curve for that system?

I have always been interested in seeing what other people can come up with when it comes to fully automated trading strategies. By that I do not mean them actually revealing their strategy logic (why would they?) but rather providing a glimpse of what risk/return/drawdown etc. their strategy can achieve. So here is what I would like to see ideally in this thread. Whoever has a strategy that is fully automated (purely mechanical), please share some of your backtested (or better yet traded) results. The only requirement I pose here: only include OUT OF SAMPLE tests of your systems. By the same token, here is what I wouldn't like to see: a bunch of people saying it's impossible to come up with a purely mechanical trading strategy. To be fair I am attaching a snapshot of one of my strategies, I hope to see many others. StrategySnapshot.xls

Did you look what Japanese stock market did in the past 30 years? Would make you think twice about your strategy. There is no guarantee that the stock market will generally go up long term.

Thanks for your response. I retested the system with limit orders on 5 min bars and it shows similar to your results. The problem I would have with this system is that it's performance before October/November 2008 (incidentally this coincides with the market crash) is the exact opposite-equity curve steadily going down. In any case, I understand your logic and wish you luck trading it until it stops working. Just a quick question here. What drawdown will cause you to stop using this system? I am asking because the equity curve going back to 2005 has a sharp V shape with turning point in 2008, so once the equity curve starts going down you cannot be confident it's just a temporary blip or the system simply doesn't work anymore. Regardless of all the ranting here, I do like your systematic approach to researching market behaviour.

I backtested your system on 5 years of data and it doesn't show good results, it does show better results starting from 2008 but still nowhere near the equity curve you are showing. I used 1 min OHLC data to test the strategy. I am not saying simple things don't work but this just seemed very suspicious to me that such a system would work and the backtest showed it wouldn't.

Did you backtest your system to know that it gives you 90% of 0.5pts winners vs 10% 1pt (or 1.5pt) losers? If yes, how many trades did your backtest (out of sample) include? A few hundred? If yes, then you can just as easily backtest partial exits. But if your current system is REALLY 90% successful it gives you an average 3.5 pts profit on 10 trades (per contract), which coupled with a high(er) frequency of trades would create a beautifully smooth equity curve. One experience I have with testing various exits, partial exits etc. is that unless you ADD a new edge (idea) to your existing system, its results won't improve significantly by just fiddling around with partial exits vs single entry exit etc. In any case, if you do have a system that does what you say, you are good to go and milk it as much as possible.