BlowFish

I am not gonna disagree with you and I could only debate from the perspective of an old dinosaur anyway Sure it works but the problems with data disposal and garbage collection are well documented. As I won't disagree I'll have a ramble instead:D Incidentally using a pre-emptive multitasking OS for a real 'real time' project is perhaps not the best approach. Sure it is possible to throw computing power at it but in my book unless you can meet processing deadlines deterministically its not really 'real time'. Like I said ....old real time dinosaur here . In a previous life (many years ago) I was a software engineer most things where essentially the same then as they are now. (I supose OOP is new but good programmers where using OOP type approaches way before it was formalised) Anyway two or three good guys will always achieve in a year what team of a hundred average guys will never achieve....ever. Even if writing to the metal (which was frequently done to ensure performance). Anyway getting vaguely back on topic, the architecture of NT leaves much to be desired and sadly (I rather like some aspects of it) until stuff gets fixed (not likely before 7.5) I don't suppose I will use it for anything 'hardcore'. Ninja is pretty, I like the chart trader and it's order management stuff however, compare it to say....Neoticker which is not pretty but fairly hardcore (handles lots of data robustly and efficiently). Horses for courses I guess but I wish Ninja was a bit more 'hardcore' but it good as far as it goes.

Both are methods to try to detect 'trade direction'. Lots of studies have been done on how efficacious these methods are (I mentioned these earlier in the thread). Broadly speaking most conclude that V@B/A is about 10% more efficacious than uptick/downtick though results do vary. None of it is precise however they did convince me that it is a 'good enough' metric to be useful.

Handy list. Looks like a fair number have an institutional slant but things are clearly improving for retail guys. Edit: The notes at the bottom of the list are handy for those that don't know the differences between types of 'broker'.

In a market that trades 23.45 / 5 Monday to Friday the only difference trading over night is liquidity. As a newb is only going to have a small position open thats a complete non issue put your stop in and sleep soundly. If you can't sleep follow Livermores advice....reduce your position to sleeping point! I am not sure where this 'conventional wisdom' came from, presumably the days (and markets) that could open limit up or limit down for days at a time. It is just not valid nowadays with the e-minis (the subject of your question). I would be wary of anyone that just out and out advises you not to trade over night especially if there prime reason is 'overnight volatility'. Absolute nonsense (though you can verify for yourself by comparing the overnight range with the daily range in excel or similar). The advice is usually from other nubs or maybe scalpers that watch every tick. To suggest that it is appropriate for everyone is crazy talk. Funnily enough when I started out the advice tended to be the opposite 'trade the swings that last 3-5 days, if you can't catch those you will be unlikely to succeed intraday'. Mind you intra day trading was considered the devils work back then. When I used to trade in and out a lot I sometimes used to mess up and not realise I wasn't flat at the end of the day. Did it once on a Friday with a DAX position and the market was shut for the weekend by the time I realised the market was shut, it was not particularly comfortable. Guess lots have been there but not being able to add up is not a reason to refrain from having an overnight position!

One more thing occurred to me after I turned the computer off last night. As price is one dimensional, vector and scalar representations will be identical. @maxima price changes are discrete sure, but there are more than enough of them to calculate acceleration. In fact in many applications (signal processing for example) the first step is to convert a continuous series into a discrete one (sampling)! The issue is not calculating acceleration (that's pretty trivial) it is using that metric to make useful trading decisions! Acceleration will absolutely jump all over the place as the market ebbs and flows so you will probably want to rersample and/or do some sort of pre or post processing to smooth it. Taking Tams idea a stage further comparing two different period momentum indicators might yield interesting results and jives better with acceleration being a second order measure.

Why do you want something beyond a =v - v(-1) / delta t? That is the acceleration. As delta t approaches zero average acceleration approaches instantaneous acceleration. As you are using discrete intervals (ticks) that's as good as it can ever get the acceleration tick to tick. Not sure what software you are using and whether it has the necessary time resolution? Oh and the difference between v and v (scalar and vector representations) also tend to disappear as the distance approaches zero. Maybe I am missing something?

velocity is change in price/change in time acceleration is change in velocity/change in time I guess tick to tick measurements would yield closest to instantaneous. You might mess with volume as a measure of mass perhaps? I don't think you need mass unless you are looking at energy or force?

I'm possibly the wrong guy to comment I understand the broad idea that strategy - tactics - tasks are hierarchical and so vary in scope. Having said that I never know where one should draw the lines, it always seems rather arbitrary. I think it is important to consistently define your terms. I guess how you get in (and out for that matter) of the market could be considered tactics.

This would certainly work but grows exponentially slower as the set of data increases. (It calls SortArray which is a bubble sort and fairly highly iterative). Not a problem if you are looking at (for example) the close of 5 minute bars (small data set) but possibly an issue if you are using say tick data (a large data set). Just offering this for consideration .

There is some good background here Selection algorithm - Wikipedia, the free encyclopedia. Are you actually talking about price data arriving in real time? That's a wee bit tricky. Trouble with EL is that it's not great for more complex data structures. You could probably sort data into an array as it arrived. Whilst not difficult a bit fiddly. Median is the mid price in a sorted set of numbers right? (not the mid price).

Update on every tick setting may be enough for your needs depending on exactly what you are doing.

Actually can I recant that? . Let me describe a senario and possible way of approaching it and ask for your comments. Lets say a weekly dataset has a positive skew and that price has moved up from the VWAP some way towards the SD1 (1/3rd or middway, whatever). We might anticipate that price would continue to SD1 the minimum expected movement. Lets Imagine the daily profile (our chosen trading set) develops a downward skew. We might choose to skip a VWAP trade against the context of the larger dataset perhaps favouring a breakout (in accord with the larger dataset) or for waiting for the daily data set to flip. Nothing really revolutionary just a way of considering the context of the bigger picture (data set). In the HUP section you described how they (HUPs) might hold up price but it seems to me they can provide more information too. Of course you have to be careful not to information overload and to remain clear what you are actually trading. From previous posts I doubt that you would use the data like that but would still be interested in your comments.

This was Eddie Toppels (author of Zen in the Markets, or Confessions of a Zen Trader) approach to trading. Enter the market (direction unimportant), if price goes X points against you, stop and reverse, repeat until 'done'. His whole thesis was that it is ego that is the main obstacle to trading successfully.

The model seems straight forward enough DugDug, a prop shop 'mash up' really. The fee is for the training. Reasonably attractive deal on the face of it.

hehe. Some of the 'issues' with .net are well documented I wonder how many programmers are aware of them let alone capable of circumnavigating them? Not Ninjas it would seem You also have to wonder about undocumented 'issues' too. When it comes to coding I am old school - write everything yourself then you know exactly how it works.

The way to make 'more money per unit time' is by growing your account and increasing size. Eventually the issue (if you want to go that far) is finding markets that will accommodate the size you need to trade to make how much you want to make. Compound growth will trump 'efficiency' every single time. (obviously you need to maintain strangle tight control of risk too). I wouldn't mention it but imvho this I think that this is actually dangerous advice because it focuses attention in completely the wrong area and fosters a way of thinking about the markets that is not particularly healthy. Eidt: I might have got hold of the wrong end of the stick However I do remember Jacks 'continuous capital extraction' or whatever he called it.

Cant imagine how people could possibly succeed without other means to support themselves while they learn. Of course just because I can't imagine it dosen't mean that some remarkable individual hasn't done it.......still can't see it though

As an aside Neoticker (in theory, I have not used NT for quite a while) would be a good option for this type of work (with its 'tick precise' technology).

Thanks, matching exactly is pretty conclusive and that alone would add weight to the whole procedure.

A lot of his less esoteric stuff is quite solid...price action, robust money management. You can read about some of his price action stuff at Fibonacci Trader - The first Multiple Time Frame Software for Traders click on FT Journal Issues at the bottom left.

The threads on market statistics certainly captured my imagination though I am sure Jerry is much better qualified to comment. Having said that the principles that he presented are equally applicable to larger data sets (weekly and monthly) as they are small ones. I have not personally traded those but they certainly appear to give 'lines' that describe price well. The last couple of threads in the series (position trading & HUP's) deal with these larger sets. I am a bit pushed for time right now but will give it some more thought and see if I can come up with any observations that might be relevant.

I always thought filtering was a cop out, a throw back to a bygone era. It is using a band aid rather than curing the wound. Put another way with robust error checking filtering should be unnecessary. I have also wondered how an electronic exchange can be designed that can generate 'bad ticks'. Sure some fat fingered human might accidentally sell 100,000 rather than 100 but that's a different matter altogether. I guess when people talk of filtering they might mean coalescing or aggregation? Filtering to me implies purposely removing data.

There is another fairly recent thread that asks the same question, you might want to have a search.

I dunno this still sounds rather blanket : I am not after a detailed test plan just a couple of sentences on how you test and with what. Which feed? I am not sure why the secrecy tbh Is it NxCore? Thompson/Reuter? Do you check that the cumulative total matches day to day? That would seem to be a fairly good 'checksum'. Do they tally exactly? Does it vary day to day? Are you using the same client software to test? Just a couple of sentences would elevate this from being a blanket statement. Edit I am not disagreeing with your conclusions (thoughwith broker supplied feeds infrastructure plays a big part) just wondered how you arrived at them

That won't speed up the crossover it will just tell you when it might happen?