Trading Market Flow / Market Structure Statistics


TopFroxx

Well-Known Member
Hey guys,

it's me again. In my other journal thread (cointegration- / pairs-trading) i already mentioned that i was thinking about starting another journal, based on computed daily PA signals.
Well, this isnt it!
Last weekend i was browsing on another forum and found a thread that sparked my interest. it was a journal of an ex institutional trader (so he says, from what i have read i believe him, but it doesnt actually matter in the end), and i thought maybe he can tell me something that could be of value. after a few minutes i realized the thread wasnt going where i thought it would go (some "insider" insight into how "the big boys" trade), but it took another, at least as valuable route. this guy basically developed a system that is based on trading market structure (he never used that term, but that is basically what it is, at least that is how i interpreted it). he talked about it only a bit and i did not really gain any new insights. BUT: i realized something. his approach had something in common with other approaches/teachings that i found to be valuable.
first, ICT (as the one person with probably the most influence on how i view the markets) teaches how important market structure is. especially daily and h4 market structure are key if you want to capture moves of 100 pips or more (not his words, my words). in this context also another ICT student that i hold in high esteem swears on the power of market structure.
second, back on BP there was a guy called relativity. he also studied the market in terms of swings and relation of swings on different timeframes. i havent studied his thread that much, but what i gained from it was that some swings happen more frequently than other things in certain situations.
third, my own personal experience tells me the same story. but because i have trust issues and dont even trust my own judgement when it comes to trading (maybe because i studied finance and behavioral biases a bit too much) i was never really sure about it. i needed some proof.
sooo, i spent the last weekend coding and evaluating. i used daily and h4 close prices (not high and low!) on a couple of different pairs. then i ran an algorithm to extract swing points.
now how do i aggregate this into some useful statistics. i thought about what kind of "figures" there are. i defined 10 different figures, each consisting of 4 swing points and 3 swings connecting the points.
an example is attached. it shows a classical bullish move, with a higher low (point 3 > point 1) and a higher high (point 4 > point 2).
each of the figures got a number from 0 to 9. this gave me a sequence of figures. this sequence can now be analyzed to get high probability swings.

this will be a journal of trades that i take according to the sequences i gathered. for trades i will be looking at daily, h4 and even h1 swings (generated by the zigzag indicator). i will be more specific when i get to future trades.
the purpose of this is for me to keep track of what kind of trades i took and the probabilities that i assigned them. after 1 or 2 months i should be able to get an idea about if at all and if so, how much this helps me to skew the probabilities in my favor.
to start things off, i am currently long UJ on the break of 104.40 last night. h1 suggests 104.80 will be hit next over 104.00 with 58% and h4 suggests this with 77%! given a RR of 1:1 (40 pip SL and TP) this trade has a positive expected value and thus is a valid trade. it is also valid to add positions on pullbacks lower, since the statistics dont cover these smaller pullbacks and the swing points on h1 and h4 are not influenced and thus the statistics (!my swing statistics! i know that it gets less likely to hit my target when price moves away) remain the same. for example i added at 104.30 this morning. to add positions on pullbacks i will also look for OTE's and key level confluences (104.20 would be nice in this example, weekly open and ~61.8% retracement of last h1 and h4 swing).

the next trades that i take will be a bit more structured and with pictures. i would like a neat journal to work with in 2 months.

if you have any questions/suggestions, you are always welcome.
 

Attachments

TopFroxx

Well-Known Member
short addition to the uj trade: set a buy stop order just now for 104.50, SL 104.20, TP 104.80. if price gets there will move SL on all other positions to 104.20 as well.
the reason for the buy stop order is the confirmation on m15. this should be the "safest" entry, with even more confirmation. will be interesting to evaluate these trades in a few months. i think it is likely that these triple confirmed trades turn out to be the ones with the highest hit ratio (keep me out of some bad trades).
edit: profit target hit
 

TopFroxx

Well-Known Member
maybe before i start with the trades another explaining post.
attached you find the 10 swing figures that i use. all of these have unique properties regarding relative highs and lows.
the second row is just the mirror image of the first row.
when generating the sequence of figures i look at "overlapping" figures. this means that the last 3 swing points of one figure and the first 3 swing points of the following figure will be the same.
one simple example. the second attachment shows 5 swing points. the first 4 form the green figure (figure number 3), and the red one is a 9. so this whole price move (the black line) equals the sequence "39".
 

Attachments

TopFroxx

Well-Known Member
next trade (already taken yesterday):
NU short, ~0.8290, initial SL 0.8346, TP 0.8226, RR ~1:1
High (in red circle) confirmed on break of red line.
figure "0" confirmed.
sequence: 32320
Probability of success: 0.7034
Positive Expected Value? => 0.7034*1-(1-0.7034)*1>0 => YES
Additional Comments: best entry would have been 0.8330 at pullback to noted KSR level with a RR of about 6:1.
Edit: Stop adjusted to 0.8330
Edit2: Stop adjusted to 0.8280, added one small position at test and rejection of 0.8280 KSR
Edit3: Target hit (well, 2 pips short, but closed). Also realized, looking at the daily, that a move to 0.8140 is likely with 61% chance. leaving 20% of my initial trade open with stop at 0.8280 to target 0.8140.
Edit4: added a small position at 0.8245. same reason as in Edit3. as long as we are below 0.8345, higher probability to hit 0.8140
Edit5: final target hit
 

Attachments

TopFroxx

Well-Known Member
next trade. i can take two different kind of trades with these stats. the potential profitability comes either mainly through high probablity of success, or through a good RR with mediocre probability of success. the NU trade belongs to category 1, whereas the following trade belongs to category 2.
EDIT: just realized that with trades of the second category the titel "trading market flow / market structure statistics" isnt really accurate anymore. my initial findings were, that there are trades of category 1. the probabilities of about 66% or even more happen only when in line with market structure. but since i realized that i can just calculate an expected value of the trade (i have a fixed TP, SL and estimated probabilities) there is at this point in time no reason to not take those trades as well. maybe time will show that these trades arent as good, but for now i will test everything that is valid regarding expected value.

EG long, ~0.8230, initial SL 0.8170, TP 0.8330, RR ~1.7:1
low of figure "8" confirmed.
sequence on h4: 84208
Probability of success: 0.4071
Positive Expected Value? => 0.4071*1.7-(1-0.4071)*1>0 => YES
Additional Comments: KSR and falling trendline reclaimed ~0.8220, will add on a pullback to 0.8200
Edit: stop adjusted to 0.8230, BE
Edit2: SL hit. this as a retracement trade instead of reversal would have worked perfectly
 

Attachments

TopFroxx

Well-Known Member
i havent talked about money management at all so far. the reason is because i didnt really think about it (for this strategy) until now.
given the probabilities, RR ratios and two categories of trades i coded a simulation to see what to expect in terms of payoff but especially drawdown.
before i get to the results, first the settings:
fraction of type 2 trades: 0.67 (i assumed this to be twice as much as the high probability trades of type 1. my impression so far is that it's a bit easier to find setups for this category)
fraction of type 1 trades: 0.33
type 2 win%: 0.37
type 1 win%: 0.55
number of trades: 250 (assuming 1 trade a day, simulating about a year)
RR type 1: 1
RR type 2: 2.25
number of simulations: 10000
risk per trade type 1: 0.04
risk per trade type 2: 0.02
given the lower win ratio of type 2 trades, by definition there will be larger drawdowns, so i will take lower risk on those trades. i tinkered a bit with the risk per trade settings, these are the ones i found optimal given personal preferences and looking at the stats that are about to follow.

attached you find the 10000 paths and the histogram of equity values at the end of each simulation path. aditionally i computed the following stats:

median of final equity: 3.16 (316% of initial deposit)
fraction of simulations in drawdown at the end of 250 trades: 0.02
median drawdown of those 2% of simulations in drawdown: 0.142
95 percent quantile of maximum drawdowns (max drawdowns of all paths, over the complete course of each path): 0.4687

using this simulation i came to the conclusion that:
1. lower the risk of type 2 trades, or even do not trade them at all (will go with lowered risk for now, want to get some forward stats first, before i dicard them completely)
2. 4% max risk is sensible. keeps maximum drawdown below 50%, even in extreme cases.
3. given the settings i can expect to about triple the account in a year (given the settings are correct, especially the probabilities)
4. after a bit of tinkering i also found out that i should and will from now on, only take type 1 trades with an expected value of 0.15 and type 2 trades with an expected value of 0.2, giving the trades some margin for error.
 

Attachments

TopFroxx

Well-Known Member
UJ long, 102.62, initial SL 101.96, TP 104.60, RR ~3:1
low of figure "8" confirmed.
sequence on h1: 84208
Probability of success: 0.38
trade type: 2
Expected Value? => 0.38*3-(1-0.38)*1>0.2 => YES
Additional Comments: just got back up above KSR level, need to hold there a bit to confirm though. once the low on h4 is confirmed (possibly around 103.20, but need to wait and see), h4 probability will also be in my favor. will be looking for a pullback to add then.
Edit: changed my view about this trade. new target is now 103.95. trade has a higher expected value this way, because of a much higher probability of success which is now 75%. this is a different type of trade. i would call this type 3. it is high probability, but only because i target "just" the recent low and not aim for a new high. this is clearly a trade against market structure, but still a high probability retracement.
now that i think about it i would call type 1 trades "continuation" trades, type 2 trades "reversal" trades, and type 3 is then "retracement". will use these labels from now on, although "reversal" might not survive for much longer, because retracement trades probably will have a higher expected value almost every time.
Edit2: added on the pullbacks earlier and on the break above 102.77. new picture attached
Edit3: SL moved to 102.17
Edit4: SL hit
 

Attachments

TopFroxx

Well-Known Member
two different trades on 1 pair accoding to two timeframes:
trade 1:
UC short, 1.1086, initial SL 1.1135, TP 1.1035, RR ~1:1
high of figure "7" confirmed.
sequence on h1: 31087
Probability of success: 0.62
trade type: continuation
Expected Value? =>0.62-0.38>0.15 => YES
Additional Comments: in line with h4 expecation => feeling confident

trade2:
UC short, 1.1086, initial SL 1.1165, TP 1.0975, RR ~1.4:1
high of figure "3" confirmed.
sequence on h4: 41313
Probability of success: 0.65
trade type: retracement
Expected Value? =>1.4*0.65-0.35>0.2 => YES
Additional Comments: in line with h1 expecation => feeling confident (but less than with the h1 trade)

Edit: the H1 entry was maybe a bit prematurely. looking at it now, the optimal entry would have been at the break of the 1.1075 low
Edit2: closed early for a 28 pip loss. will reenter via sell stop order at 1.1075
Edit3: sell stop order also cancelled, overall a loss trade
 

Attachments

jack

Administrator
Staff member
How are you, or are you, factoring in possible hindsight bias when reviewing chart setups in your sample size for testing?

I mean, there's going to be an errors rate that can negatively impact expectancy if you're manually identifying chart patterns, even with something like a zigzag indicator picking your highs and lows.

Maybe it's small, but still.

(I ask since in an earlier post you mentioned an 'assumption' of 2/3rds of one type over another appearing or being taken.. and that got me thinking if there's an assumption of manual error as well in the same identification.)
 

TopFroxx

Well-Known Member
Jack said:
How are you, or are you, factoring in possible hindsight bias when reviewing chart setups in your sample size for testing?

I mean, there's going to be an errors rate that can negatively impact expectancy if you're manually identifying chart patterns, even with something like a zigzag indicator picking your highs and lows.

Maybe it's small, but still.

(I ask since in an earlier post you mentioned an 'assumption' of 2/3rds of one type over another appearing or being taken.. and that got me thinking if there's an assumption of manual error as well in the same identification.)
thanks for the feedback jack.
could you give me an example? i am not sure if i understand what you mean. the in sample stats are computed and the forward test is live, so i dont see where there can be a hindsight bias. manually identifying the swing figures doesnt really generate errors. i have them clearly memorized by now and given the swing points by the zigzag i can generate a sequence of 5 figures within a few seconds. also on my charts (as you can see) i have always written down the recent sequence and just need to add one or two new numbers once the swings form.
and as a general margin for error i made the rule "expectancy > 0.15" or ">0.2" for type 1 and type 2 trades, respectively.
 

jack

Administrator
Staff member
TopFroxx said:
thanks for the feedback jack.
could you give me an example? i am not sure if i understand what you mean. the in sample stats are computed and the forward test is live, so i dont see where there can be a hindsight bias. manually identifying the swing figures doesnt really generate errors. i have them clearly memorized by now and given the swing points by the zigzag i can generate a sequence of 5 figures within a few seconds. also on my charts (as you can see) i have always written down the recent sequence and just need to add one or two new numbers once the swings form.
and as a general margin for error i made the rule "expectancy > 0.15" or ">0.2" for type 1 and type 2 trades, respectively.
My bad then, I got the impression you were manually identifying trade setups and triggers, and just running stats on the manually collected sample of backtested trades. If your hit rate and frequency of type 1 & 2 setups is collected from historical data the same way new trades are taken automatically, then there's no issues with human error to factor in..

If there was, then the human element of hindsight bias makes it harder to 'see' failed setups in past data as you scroll through the charts.. our eye simply finds it harder to pick out the bad trades given we'd be unconsciously looking for the positive follow through to make a setup stand out.

(Hope that makes sense, I'm half asleep right now.)
 

TopFroxx

Well-Known Member
Jack said:
My bad then, I got the impression you were manually identifying trade setups and triggers, and just running stats on the manually collected sample of backtested trades. If your hit rate and frequency of type 1 & 2 setups is collected from historical data the same way new trades are taken automatically, then there's no issues with human error to factor in..

If there was, then the human element of hindsight bias makes it harder to 'see' failed setups in past data as you scroll through the charts.. our eye simply finds it harder to pick out the bad trades given we'd be unconsciously looking for the positive follow through to make a setup stand out.

(Hope that makes sense, I'm half asleep right now.)
i think i know what you mean. but in this case there is no hindsight bias involved. matlab has done the backtesting und computation of probabilities. new trades are taken manually though, but there is no hindsight bias in forward testing.
 

TopFroxx

Well-Known Member
EA long, 1.5605, initial SL 1.5495, TP 1.5750, RR ~1.3:1
figure 1 confirmed
sequence on h4: 84231
Probability of success: 0.68
trade type: continuation
Expected Value? =>1.3*0.68-0.32>0.15 => YES
Additional Comments: bounce from old high and KSR level, h1 confirmation, in line with overall trend, confident, classic continuation trade in line with HTF market structure
Edit: SL hit
 

Attachments

TopFroxx

Well-Known Member
EN long, first entry on pullback to old high at 1.6480, initial SL 1.6260, TP 1.6870, RR ~1.8:1
figure 2 confirmed
sequence on D1: 84232
Probability of success: 0.55
trade type: continuation
Expected Value? =>1.8*0.55-0.45>0.15 => YES
Additional Comments: added positions on break of 1.6560, h4 confirmed and SL adjusted to 1.6470, first trade on the daily timeframe with h4 confirmation now.
Edit: took partial profit at 1.67. SL adjusted to 1.6570
edit2: had SL adjusted to 1.6595, got taken out
 

Attachments

TopFroxx

Well-Known Member
two trades i already took on thursday:

UCHF long, 0.9015, initial SL 0.8945, TP 0.9120, RR ~1.5:1
figure 5 confirmed
sequence on D1: 84975
Probability of success: 0.73
trade type: continuation
Expected Value? =>1.5*0.73-0.27>0.15 => YES
edit: SL adjusted on 50% to 0.9020
edit2: 50% taken out at 0.9020


GU short, 1.6480, initial SL 1.6560, TP 1.6330, RR ~1.9:1
figure 3 confirmed
sequence on D1: 41323
Probability of success: 0.55
trade type: reversal
Expected Value? =>1.9*0.55-0.45>0.2 => YES
edit: target hit
 

Attachments

TopFroxx

Well-Known Member
UCAD long, 1.11, initial SL 1.1046, TP 1.1196, RR ~1.8:1
figure 2 confirmed
sequence on H4: 13132
Probability of success: 0.56
trade type: continuation
Expected Value? =>1.8*0.56-0.44>0.15 => YES
comments: entered the first position a bit prematurely, before the break of 1.11. have a buy stop order set at 1.1127, will be a 1:1 RR trade for that position. i'd like to get into the habit of averaging up.
edit: adjusted the sl to 1.1057 during the day yesterday and got hit
 

Attachments

Tansen

Well-Known Member
TopFroxx said:
UCAD long, 1.11, initial SL 1.1046, TP 1.1196, RR ~1.8:1
figure 2 confirmed
sequence on H4: 13132
Probability of success: 0.56
trade type: continuation
Expected Value? =>1.8*0.56-0.44>0.15 => YES
comments: entered the first position a bit prematurely, before the break of 1.11. have a buy stop order set at 1.1127, will be a 1:1 RR trade for that position. i'd like to get into the habit of averaging up.

Don't let this influence your original decision but, I remember Lori tweeted something about the banks calling for 1.18 in USD/CAD, whether or not that's going to happen is something only time can tell.

http://business.financialpost.com/2014/01/23/3-reasons-the-canadian-dollar-will-shed-another-10-15/

Link he posted and it was a tweet on the 27th of Jan
 

TopFroxx

Well-Known Member
Tansen said:
Don't let this influence your original decision but, I remember Lori tweeted something about the banks calling for 1.18 in USD/CAD, whether or not that's going to happen is something only time can tell.

http://business.financialpost.com/2014/01/23/3-reasons-the-canadian-dollar-will-shed-another-10-15/

Link he posted and it was a tweet on the 27th of Jan
i am already in the trade and also it's in the right direction. just further confirmation if anything. but thanks for the input
 

TopFroxx

Well-Known Member
sorry for the lack of updates. will post a couple of trades from the past 1 or 2 weeks soon. didnt have the time to trade much let alone to journal it here
 

Top