Synthetic hedges, cointegration, mean reversion and similar stuff

Quoting Ultimate6

Disliked

Cool. Yea USDX chart is always open on my screens when trading.

Ignored

Quoting Dirtybrown

Disliked

So when do you think you would be the best time to book profit on this hedge? It's been fluctuating mostly within $100-$150 profit. Would it be when the bottom channel line on the R indicator was touched?

Or could i just keep it and repeat the same trade tomorrow.. keep both then cloe all at the very end of the week? or is that missing the point. I just want to put soem kind of spin on this to book profit regularly then re-enter hedges at the right time.

Ignored

Quoting Ultimate6

Disliked

Looks good. I'm still too caught up in theory to apply practise. That looks like a trendslope chart. Are you sure you haven't put some R code from the trend ea into the arb one?

Ignored

Quoting Dirtybrown

Disliked

Hey, actually I think you were right, somehow it was showing it as a trend, maybe I flipped the setting and it didn't take it back off, who knows. After restarting the EA 3 times it seems to display normal. Getting these proportions this morning which i just entered on another demo


Attachment

Ignored

Quoting Dirtybrown

Disliked

Hey, actually I think you were right, somehow it was showing it as a trend, maybe I flipped the setting and it didn't take it back off, who knows. After restarting the EA 3 times it seems to display normal. Getting these proportions this morning which i just entered on another demo


Attachment

Ignored

Quoting FXEZ

Disliked

Dirtybrown,
Notice how different the eigen values are between the two latest pictures you posted. I've found that if you play around with arb-O-mat you can get it to produce very trend-like curves in some cases, simply due to the regression. I'm sure the initial thinking behind trend-O-mat was that if you have an upward sloping trend in addition to the standard deviation lines for timing of entry and exit, you have an advantage of the trend working in your favor if you just favor buys.

Unfortunately trend-O-mat seems to be based on a spurious...

Ignored

Quoting Dirtybrown

Disliked

Hey there, thanks for the message I've read many of your posts quite insightful. Yeah, I couldn't make out how to interpret the Trendo Mat whatsoever, so sticking to the Arb. Given the above, when would you say would be the right time to enter the hedge? At the mean (blue center dashed line) or at the within the, say, .004 - 002 range?

Something else to note: Prior to London open, I notied the basket abovei in profit $300, which is the highest I've seen. Being this is just an experiment, I left it open and it is now currently negative.

Rick...

Ignored

Quoting Ultimate6

Disliked

Yea even though I'm currently testing on FX, I think there is greater potential in other markets and/or mixed products

Ignored

% make sure previously defined variables are erased.
clear;
% read a spreadsheet named "GLD.xls" into MATLAB.
[num, txt]=xlsread('GLD');

% the first column (starting from the second row) is
% the trading days in format mm/dd/yyyy.
tday1=txt(2:end, 1);
% convert the format into yyyymmdd.
tday1=..
datestr(datenum(tday1, 'mm/dd/yyyy'), 'yyyymmdd');

% convert the date strings first into cell arrays and
% then into numeric format.
tday1=str2double(cellstr(tday1));
% the last column contains the adjusted close prices.
adjcls1=num(:, end);
% read a spreadsheet named "GDX.xls" into MATLAB.
[num2, txt2]=xlsread('GDX');
% the first column (starting from the second row) is
% the trading days in format mm/dd/yyyy.
tday2=txt2(2:end, 1);
% convert the format into yyyymmdd.
tday2=..
datestr(datenum(tday2, 'mm/dd/yyyy'), 'yyyymmdd');

% convert the date strings first into cell arrays and
% then into numeric format.
tday2=str2double(cellstr(tday2));
adjcls2=num2(:, end);

% find all the days when either GLD or GDX has data.
tday=union(tday1, tday2);
[foo idx idx1]=intersect(tday, tday1);

% combining the two price series
adjcls=NaN(length(tday), 2);
adjcls(idx, 1)=adjcls1(idx1);

[foo idx idx2]=intersect(tday, tday2);

adjcls(idx, 2)=adjcls2(idx2);

% days where any one price is missing
baddata=find(any(~isfinite(adjcls), 2));
tday(baddata)=[];

adjcls(baddata,:)=[];

vnames=strvcat('GLD', 'GDX');

% run cointegration check using
% augmented Dickey-Fuller test
res=cadf(adjcls(:, 1), adjcls(:, 2), 0, 1);
prt(res, vnames);

% Output from cadf function:

%  Augmented DF test for co-integration variables:
   GLD,GDX
% CADF t-statistic        # of lags       AR(1) estimate
%      −3.35698533                1            −0.060892
%
%    1% Crit Value    5% Crit Value       10% Crit Value
%           −3.819           −3.343               −3.042

% The t-statistic of −3.36 which is in between the
% 1% Crit Value of −3.819
% and the 5% Crit Value of −3.343 means that
% there is a better than 95%
% probability that these 2 time series are
% cointegrated.

  


results=ols(adjcls(:, 1), adjcls(:, 2));

hedgeRatio=results.beta
z=results.resid;

% A hedgeRatio of 1.6766 was found.
% I.e. GLD=1.6766*GDX + z, where z can be
% interpreted as the
% spread GLD-1.6766*GDX and should be stationary.

% This should produce a chart similar to Figure 7.4.
plot(z);

Quoting Dirtybrown

Disliked

Spent more time doing research this weekend. For any computer nerds out there. Oreilly's Safari offers an e-bookshelf service that let's you pick out 5 books for only $10/mo. One of them I got was Quantitative Trading: How to Build Your Own Algorithmic Trading Business by Ernest Chan. There are lots of nice matlab example codes throughout. This is how he tests for cointegration in matlab. It also gives the hedge ratio:

I'm trying this with usdcad / nzdusd first.

Ignored

Quoting Ultimate6

Disliked

Are you thinking of using Matlab instead? I would hope R gives the exact same results

Ignored

Quoting Dirtybrown

Disliked

Hey there, thanks for the message I've read many of your posts quite insightful. Yeah, I couldn't make out how to interpret the Trendo Mat whatsoever, so sticking to the Arb. Given the above, when would you say would be the right time to enter the hedge? At the mean (blue center dashed line) or at the within the, say, .004 - 002 range?

Something else to note: Prior to London open, I notied the basket abovei in profit $300, which is the highest I've seen. Being this is just an experiment, I left it open and it is now currently negative.

Rick...

Ignored

Quoting FXEZ

Disliked

It seems to me that if the pairs are really cointegrated then mean reversion will be the best bet. You can do it in a number of ways, and the results from testing the different alternatives may make the decision more clear for this set of symbols. Cost should also be taken into effect.

If spread cost is 2 pips per pair * 7 pairs traded then 14 pips must be overcome to make a profit. In many cases the cost will be less than 14 pips due to the bid/ask shifts of the pairs at the time of entry/exit, but composite spread for all 7 pairs should be...

Ignored

Quoting marijan993

Disliked

Hi guys, were can I find this arb-o-mat indicator? I'm not sure I understood this strategy reading your posts. If I could look at some indicators on my charts maybe I could figure something out. Thanks.

Ignored

Quoting Dirtybrown

Disliked

Please read through the first few posts. 7bit wrote a custom dll to send metatrader data to a free statistical package R which does the calculations.

Shafiq, sure, send what you have.

Ignored