Tick Statistics | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Tick Size, Tick Rate, Tick Spread, and Tick Logger indicators 36 replies

Tick by Tick trading? 5 replies

To Tick or not to Tick? (any strategy here?) 7 replies

download data tick by tick ??? help 1 reply

Attachments: Tick Statistics
Exit Attachments

Tick Statistics

  • #1
  • First Post: Sep 18, 2010 2:16am Sep 18, 2010 2:16am
Hey Guys,

I've searched and searched but have not found an answer to my simple question. This is for you stat guys out there. We all know that market can do one of three things, tick up, tick down, tick sideways. So here is the question, after say 10K ticks (insert your own sample size), would we expect to see an equal distribution of those three possibilities?

Thanks,

Jim (scotty)
  • #2
  • Edited 6:25am Sep 18, 2010 6:14am | Edited 6:25am
Hi Scotty....

No, there is no equal distribution of those three possibilities,
and I can guarantee you that!!!
If you wish, I can send you a code for that and you can see it yourself.
(I think you already know my mail)

My code will show you exact number of up vs down ticks in every bar,
and it will count all of those ticks (ask tick, bid tick, ask tick/bid static,
and bid tick/ask static) bassicaly there is 4 possibilities....all of that info is cumulative and it gives you exact count of these scenarios via text info!

ps...try to check ticks just before news time

edit; it is most important to find suitable broker for this....you will need "real" feed aka bank/tier1 feed....just observe the ask/bid ticks in your mt4 trade screen...
if your brokers ask and bid are always moving the same way(sintetic pricing)....those are crap feeds and I personaly run away from brokers that operate that way!!!!

Regards
  • #3
  • Sep 18, 2010 6:41am Sep 18, 2010 6:41am
Quoting Parazite
Hi Scotty....

No, there is no equal distribution of those three possibilities,
and I can guarantee you that!!!
If you wish, I can send you a code for that and you can see it yourself.
(I think you already know my mail)

My code will show you exact number of up vs down ticks in every bar,
and it will count all of those ticks (ask tick, bid tick, ask tick/bid static,
and bid tick/ask static) bassicaly there is 4 possibilities....all of that info is cumulative and it gives you exact count of these scenarios via text info!

ps...try to check ticks...
Long time no see, my friend. Thanks for your response. I've been working in excel studying random number sequences (both excel generated and random.org generated) and have found that it's not that hard to predict near term fluctuations with some scotty b elementry school stat methods. I've also figured out that even tiny biases to one side of possible 'flow' create up trends or down trends over many data points that are generated randomly.

I have not yet applied what I've been working on to actual tick data, but then again, I already know that the market isn't totally random. One could argue however that even rational traders in a non random market would casue an even distribution with their orders over time. Say for example that I sell EU because of an adverse change in interest rates. My trade might be informed, but the interest rate change could be called random. I'm gonna shutup now for fear that I'm not making any sense.

I'd be interested in seeing your code. I'll PM you my email.

Thanks again,

Jim
  • #4
  • Sep 18, 2010 7:38am Sep 18, 2010 7:38am
Quoting Scotty B
I've been working in excel studying random number sequences (both excel generated and random.org generated) and have found that it's not that hard to predict near term fluctuations with some scotty b elementry school stat methods.
Random number sequences are by definition not predictable.

I've seen studies which show a powerful negative first-lag autocorrelation of ticks (after an up tick a down tick is more likely), but I didn't see any stats on tick distributions in a sample period.

http://dl.dropbox.com/u/190212/ff/first_lag.png
  • #5
  • Sep 18, 2010 8:10am Sep 18, 2010 8:10am
Quoting Adal
Random number sequences are by definition not predictable.

I've seen studies which show a powerful negative first-lag autocorrelation of ticks (after an up tick a down tick is more likely), but I didn't see any stats on tick distributions in a sample period.
Hey Adal,

Thanks for stopping by. Shamefully, I must admit that I have almost no background in statistics other than what I've been learning on my own time over the last couple years. I don't even know what an auto correlation is. Sure, I'll wikipedia it, but I apologize for not being able to engage you yet on what you've shared.

My current and albeit math deficient thinking is that when you flip a fair coin, you're probably not going to get 1000 heads in a row. That could happen I suppose, but it's probably not very likely. If it were to happen, I'd look at the event like we look at bubbles in the world markets. A pressure builds on the other side of the coin as it were, and a rush of tails would follow in subsequent flips. It might not be a move of continual tails, but by the time you got to the 2000 flip mark, you'd expect to have close to 1000 of both heads and tails. Because the 1000 heads is a pretty rare event, it might take until 10K or more flips for the distribution to equalize. So basically it is the running distribution imbalance that makes a prediction possible. It is not a single event prediction( such as predicting the very next flip), but a prediction based on the fact that the distribution over time will equalize.

If you have a situation where you are net positive heads and negative tails at the 5000th flip for example, you could expect less heads and more tails in the flips to come.

I don't know enough yet to know if this info is usable, but it seems promising. You'd need a massive series of flips to determine what an average deviation from a running distribution is, but if there was such a thing it could be very helpful over time.

All I'm doing is getting random numbers between -1 and 1 and summing the 'flow' to produce a price like chart. I've then setup a running calculation to show the variance from the expected distribution per period and it seems to be a very useful indicator. I'll post some screenshots.

Thanks again,

Jim
Attached Image(s) (click to enlarge)
Click to Enlarge Name: random price.JPG Size: 77 KB
Click to Enlarge Name: variance.JPG Size: 96 KB
  • #6
  • Sep 18, 2010 10:24am Sep 18, 2010 10:24am
Quoting Scotty B
My current and albeit math deficient thinking is that when you flip a fair coin, you're probably not going to get 1000 heads in a row. That could happen I suppose, but it's probably not very likely. If it were to happen, I'd look at the event like we look at bubbles in the world markets. A pressure builds on the other side of the coin as it were, and a rush of tails would follow in subsequent flips. It might not be a move of continual tails, but by the time you got to the 2000 flip mark, you'd expect to have close to 1000 of both heads and tails....
Mathematically speaking, if the series of flip coins is truly random, after the 1000 same side flips, you could have a billion more on the same side. You can't make any prediction at all based on passed events if the series is random and each flip is independent of the others. This is the pure math.

But the real world is impure: http://comptop.stanford.edu/preprints/heads.pdf

What you are talking about is some sort of mean-reversion. I'm not sure how you would bet on that. Let's assume that you know that in the next 10K ticks, there will be more up-ticks than down-ticks. How do you play that? You can't just go long, since the down-ticks could move more than the up-ticks. I also don't see how that would impact the volatility, which you could bet on through options.
  • #7
  • Sep 18, 2010 10:37am Sep 18, 2010 10:37am
  •  Czech
  • | Joined Feb 2010 | Status: You Don't Know What You Don't Know | 111 Posts
Hey Jim, interesting stuff. The coin-flip scenario I'd outright disagree with. As far as i can tell this is the "Gamblers Fallacy" which is "the belief that if deviations from expected behaviour are observed in repeated independent trials of some random process then these deviations are likely to be evened out by opposite deviations in the future"

Based on that i would expect real random numbers to fail. But i'm not a stats guy either so i could be completely wrong here.

We know the markets aren't random so maybe they do follow a distribution. However, I believe the market is reacting to new events all the time so i would expect that we would not see an equal distribution of up, down and sideways ticks BUT that the expected distribution would be changing constantly.

Still, I'd love to see an indicator such as Parazite's.
  • #8
  • Sep 18, 2010 10:44am Sep 18, 2010 10:44am
Interesting thread - yes a "coin" has no memory imo.
Time hides Nothing
  • #9
  • Sep 18, 2010 10:59am Sep 18, 2010 10:59am
Quoting Adal
Mathematically speaking, if the series of flip coins is truly random, after the 1000 same side flips, you could have a billion more on the same side. You can't make any prediction at all based on passed events if the series is random and each flip is independent of the others. This is the pure math.

But the real world is impure: http://comptop.stanford.edu/preprints/heads.pdf

What you are talking about is some sort of mean-reversion. I'm not sure how you would bet on that. Let's assume that you know that in the next 10K ticks, there...
I think I see what you're saying, but because there are two possibilities with each flip, you can track the distribution each period. That is what I'm doing with my excel 'indicator.' If there are more heads than expected at flip 5000 for example, and less tails I'd expect the 'price' on my running sum chart to fall, and that's what usually happens in those situations, simply because the distribution of heads and tails begins balancing. I've been doing runs of only 30K flips and I've been finding that a 4% spread, or unbalance develops at the most between the two possibilities. When that 4% mark is reached, if heads are on top I sell, if tails, I buy. To 'trade' this though, you'd need more data to get a better idea of the characteristics of the spreads that develop.

Here is a 30K flip run with a % variance from the expected distribution along with the price chart and a chart of all three heads/tails variance with the expected period distribution. You can see much on the latter because there are so many data points, but you'll get the idea.

Early on the % variance chart you can see that there is a deficiency in tails (pink line) and a surplus of heads. A down trend develops shortly thereafter.
Attached Image(s) (click to enlarge)
Click to Enlarge Name: randprice.JPG Size: 94 KB
Click to Enlarge Name: Randpercent.JPG Size: 81 KB
Click to Enlarge Name: distrun.JPG Size: 71 KB
  • #10
  • Sep 18, 2010 11:09am Sep 18, 2010 11:09am
Quoting CindyXXXX
Interesting thread - yes a "coin" has no memory imo.
Yes, I agree, but as you keep flipping and your collection of flips gets bigger and bigger, temporary imbalances will disappear and you'll have roughly as many heads as you do tails. Imbalances happen all the time. When you have a series of more heads, or +1's, the running sum or 'price' spikes. When the unbalance becomes great enough, you start getting more tails again (-1) and less heads during the re-balancing and the running sum falls.

The % variance chart I just posted does not develop anywhere near a 4% spread, 4% is the biggest I've seen so far. On that chart, if it's not clear already, if the value is above 0 it's that many percent above the current expected distribution and vise verse below the zero line.
  • #11
  • Sep 18, 2010 11:24am Sep 18, 2010 11:24am
Quoting Scotty B
Yes, I agree, but as you keep flipping and your collection of flips gets bigger and bigger, temporary imbalances will disappear and you'll have roughly as many heads as you do tails. Imbalances happen all the time. When you have a series of more heads, or +1's, the running sum or 'price' spikes. When the unbalance becomes great enough, you start getting more tails again (-1) and less heads during the re-balancing and the running sum falls.
Yes thats the interesting bit Scotty - always enjoy your posts

Theres a trader around here called amyc she trades this way - (I think) always trading back to a mean - its a very original style compared to the others around here. Not sure if its the same thing as what you're talking about but reminds very much of it. She has a journal on here where she documents here method. Its interesting anyway - she does quite well too.
Time hides Nothing
  • #12
  • Sep 18, 2010 11:35am Sep 18, 2010 11:35am
Could it be though the same way an average smooths out closer to a mean the longer the period?
Time hides Nothing
  • #13
  • Sep 18, 2010 12:23pm Sep 18, 2010 12:23pm
I think it would be better to run these tests with real tick data.

The tick data from here (http://www.truefx.com/dev/data/2010/AUGUST-2010/) has for example 30K ticks in a EUR/USD 4 hour period (depends of time of day).

That sounds to me like a too little sample. What if those 4 hours were during a massive selling of some currency? It would be like buying after a big down H4 candle. That's not guaranteed to bring a profit.

But, one HFT guy mentioned in an interview doing stuff which sounds like this - buying after 4 down ticks for example. So it might be worth it to investigate more.

Anyway, I'm getting close to finishing my own tick-based backtester. I will post some "tick imbalance indicator" stats once I'm done.
  • #14
  • Sep 18, 2010 2:35pm Sep 18, 2010 2:35pm
There's one thing you need to know about the random walk, and that is that the current state is the best estimation of the future state. Remember that the random walk has no memory!

Relating this to your example: if you followed the outcomes of a coin toss, and at one point you have counted a surplus of a 1000 heads, the best estimate that you can make at that point in time, is that this 1000 level will become the new mean. This means that once this point has been reached, the likelihood of reversion to the initial mean (which was 0 when you started counting) is just as likely as a continuation to a surplus of 2000 heads.

Whether this holds in the real world is up to you to research, but from a theoretical perspective dealing with real random walks, you're on the wrong path.

On a side note, in forex price distributions will not follow an ordinary random walk, though it might seem so if you study tick data. Instead, the expected change in the exchange rate reflects the interest rate differential of the base currency and the variable currency. This is based on the concept of interest rate parity, which has been shown to hold in the long run (lets say 20+ years). Despite the long horizon, the expected exchange rate of tomorrow, the next hour, or the next tick, is affected by this interest rate differential. Keep in mind that this mean is extremely small (but from a theoretical perspective not negligible) for short timeframes.
The nail that sticks out gets hammered back in
  • #15
  • Sep 19, 2010 2:16am Sep 19, 2010 2:16am
Quoting Noloqy
There's one thing you need to know about the random walk, and that is that the current state is the best estimation of the future state. Remember that the random walk has no memory!

Relating this to your example: if you followed the outcomes of a coin toss, and at one point you have counted a surplus of a 1000 heads, the best estimate that you can make at that point in time, is that this 1000 level will become the new mean. This means that once this point has been reached, the likelihood of reversion to the initial mean (which was...
Thanks, Nology. I suppose I just don't understand the idea. Maybe I'm leaning too much on the idea that in a given sample one should expect to see around 50% of both heads and tails. So after the thousand consecutive run of heads, our 'price' chart would show a unidirectional spike of +1000 points. Is it wrong to assume that by the 5000th flip that there might be an even distribution again? I suppose the problem of prediction just shifts from what will the price do, to when will the distribution equalize. So intead of saying that in between the 1001th flip and the 5000th flip we can expect to see 1500 more heads and 2500 tails, it is actually impossible to say when the distribution will equalize-- it might take until the 20k flip. Still a perplexing problem though.
  • #16
  • Sep 19, 2010 11:46am Sep 19, 2010 11:46am
Quoting Parazite
Hi Scotty....

No, there is no equal distribution of those three possibilities,
and I can guarantee you that!!!
If you wish, I can send you a code for that and you can see it yourself.
(I think you already know my mail)

My code will show you exact number of up vs down ticks in every bar,
and it will count all of those ticks (ask tick, bid tick, ask tick/bid static,
and bid tick/ask static) bassicaly there is 4 possibilities....all of that info is cumulative and it gives you exact count of these scenarios via text info!

ps...try to...
I never checked the statistical behaviour of price ticks but IMHO the number of up ticks & down ticks in any single bar may or may not distributed equally because a single bar itself has a bias. Logically the number of up ticks should be more than down ticks in a bullish bar.

However if we choose a large sample of bars and count the total number of up ticks and down ticks then they should be approaching 50-50.
  • #17
  • Sep 19, 2010 11:46am Sep 19, 2010 11:46am
Quoting Scotty B
Yes, I agree, but as you keep flipping and your collection of flips gets bigger and bigger, temporary imbalances will disappear and you'll have roughly as many heads as you do tails. Imbalances happen all the time. When you have a series of more heads, or +1's, the running sum or 'price' spikes. When the unbalance becomes great enough, you start getting more tails again (-1) and less heads during the re-balancing and the running sum falls....
The reason that you found max 4% spread for 30k flips because number of sample size will be increased exponentially for a lower spread and eventually become infinite for a zero spread. Just look at the details from wikipedia.
http://upload.wikimedia.org/wikipedi...ge_numbers.gif
  • #18
  • Sep 19, 2010 12:14pm Sep 19, 2010 12:14pm
Quoting Adal
Mathematically speaking, if the series of flip coins is truly random, after the 1000 same side flips, you could have a billion more on the same side. You can't make any prediction at all based on passed events if the series is random and each flip is independent of the others. This is the pure math.

But the real world is impure: http://comptop.stanford.edu/preprints/heads.pdf

What you are talking about is some sort of mean-reversion. I'm not sure how you would bet on that. Let's assume that you know that in the next 10K ticks, there...
If i have a 1000 consecutive same side flips, I will borrow as much as I can to bet that the next flip will be same again. Why? Because the both side of the coin are made of head or tail.
  • #19
  • Sep 19, 2010 5:48pm Sep 19, 2010 5:48pm
Quoting ForexQuant
The reason that you found max 4% spread for 30k flips because number of sample size will be increased exponentially for a lower spread and eventually become infinite for a zero spread. Just look at the details from wikipedia.
http://upload.wikimedia.org/wikipedi...ge_numbers.gif
Hey Quant,

Thank you for sharing. I was not aware of this law specifically, but practically I could see that the most indicative portion of my spread chart came somewhat close to the 30% region of my chart. After long horizons have been met, the data becomes less useful. It would make more sense to reset the starting point of the running spread calculations at a regular interval. My sense is that to trade in this sort of way, if tick data in fact behaves similarly, you would need lots of historical data to develop a profile of specific pair behavior to know what average maximum spreads were on average between up ticks and down ticks from what would be expected in a given sample size.

I agree with you on the market usually having a bias. As I mentioned earlier, even a tiny strength bias to one side of the market will push it in the direction of the bias over time. Using several different random number generators, I took several sets of 10K data points where I asked for numbers between (-999,1000). Over time, that extra 1 point that is generated when 1000 comes up on the positive side, skews the moving sum into a down trend. As you adjust the numbers and give one side more or less power, you can produce nice clean straight trends, or general trends with 'heavy' pull back periods.

In real life, I am in agreement with you. Apart from speculative interests and given relative economic and political stability, the market will be incentivized on one side because of interest rate differentials. Just look at AU, NZU, TRYJPY, just some examples. Where there is an incentive to carry, you will usually see this bias born out in the price data over time. At which time horizons that bias can be seen is what I'm interested in.

Thanks again,

Jim
  • #20
  • Oct 16, 2010 9:24am Oct 16, 2010 9:24am
I stumbled across this thread and was glad to read your finding.

I did notice the coin flip evident in the charts.

Based purely on candle colour.

I backtested different time frames yearly, monthly, weekly, daily, right down to 5 minute.

And regardless of trend the biggest deviation from 50/50 was 56/44 on the weekly.

because the average true range indicator was based on high /low not open/ close,I couldn't work out accurately the risk reward.
And i can only based my trades on open /closes.


But i can say for certain coin toss data is evident in candle colours.
Good luck I hope to find a way to expoit this.
making green pips large and red pips small this is my goal.
0 traders viewing now