Technical Indicators

A single function that calculates 27 different technical indicators

Description

INDICATORS is a technical analysis tool that calculates various technical indicators. Technical analysis is the forecasting of future financial price movements based on an examination of past price movements. Most technical indicators require at least 1 variable argument. If these arguments are not supplied, default values are used.

The included indicators are:

Momentum:
    Commodity Channel Index
    Rate of Change
    Relative Strength Index
    Fast Stochastic Oscillator
    Slow Stochastic Oscillator
    KDJ Indicator
    William's %R
    Aroon
    True Strength Index
Trend:
    Simple Moving Average
    Exponential Moving Average
    Moving Average Convergence Divergence
    Wildmer's DMI (ADX)
    T3
Volume:
    On-Balance Volume
    Chaikin Money Flow
    Force Index
    Money Flow Index
Volatility:
    Bollinger Bands
    Keltner Channels
    Average True Range
    Volatility Ratio
    Highest High, Lowest Low
Other:
    ZigZag
    Price Comparison
    Pivot Points
    Stop and Reverse

Version : 1.1.3 (05/24/2013)
Author : Nate Jensen
Created : 10/10/2011
History :
 - v1.0 10/25/2011 : initial release of 21 indicators
 - v1.1 03/04/2012 : 23 indicators, fixed date conversion issue
 - v1.1.1 03/25/2012 : 24 indicators
 - v1.1.2 03/21/2013 : 25 indicators
 - v1.1.3 05/24/2013 : 27 indicators, bug fixes

• MATLAB release: MATLAB 7.14 (R2012a)

[Jose]

Nate,
Thank you very much.

[Nate Jensen]

The issues have been fixed and the update should appear shortly. I have also added 2 more indicators.

The example should have read:
load disney.mat
vout = indicators([dis_HIGH,dis_LOW,dis_CLOSE],'fsto',14,3);
fpctk = vout(:,1);
fpctd = vout(:,2);
plot(1:length(fpctk),fpctk,'b',1:length(fpctd),fpctd,'g')
title('Fast Stochastics for Disney')

There was also an error in the FSTO and SSTO functions.

[Nate Jensen]

Jose,

Thank you for pointing this issue out! I will upload a fix tonight.

I'm sorry about the error.

Nate

[Jose]

I do not understand what is happening. I'm using your example!!!

load disney.mat
[fpctk,fpctd] = indicators([dis_HIGH,dis_LOW,dis_CLOSE],'fsto',14,3)

Error using indicators
Too many output arguments.

[Nate Jensen]

Ha,

I have just added the KDJ Indicator as requested. It should appear shortly depending on how long Matlab takes to approve my submission.

This indicator was a little confusing. I'm not sure exactly which method is correct so I input 3 different methods. Simply uncomment whichever method you wish to use.

Method # 1:
J = 3*FastK - 2*FastD;

Method # 2:
J = ema(K,period);

Method # 3:
J = 3*SlowK - 2*SlowD;

Hope that helps!

Nate

[Ha]

could u add KDJ to the next version?

[Nate Jensen]

Thanks Lu Li!

I'll try my hand at the mex function, but unfortunately I don't have a whole lot of experience with them so no guarantees.

I do agree with you, this function is slow as an optimizer. I suggest pulling out the individual indicators that you are interested in and turn them into their own functions. Then try to optimize those, that way you aren't bogged down by the 'Switch' statement.

[lu li]

Excellent work!
I want to use it in an optimization, but it is kind of slow.
It would be wonderful if it can be into mex funtcion so the it will be much faster.

[Nate Jensen]

Any time! Thanks again.

[Arty]

Works like a charm. Thanks.

[Nate Jensen]

Arty - If you look at line 680,
ema6 = [nan(5*period-1,1); ema(ema5(~isnan(ema5)),period,observ-5*period+1)];
notice that ema6 must have a minimum of '5*period-1' for the size of the input vector. In your case you input a period of 20, therefore, you would need at least 5*20-1 (or 99) for size of your input vector. Either increase the size of your input vector, or decrease the period.
Thanks for the compliment!

Peter - Yeah there are many calculations like that in my code where you could use one method or another to calculate ema or use sma over ema. It is really a matter of opinion. Thank you for the rating!

[Peter]

Noticed the stochastics delay is calculated using EMA, my TA program calculates it using SMA so there was a slight discrepancy.

[Arty]

Hi Nate, I tried the t3 indicator and got an error message:

>> t3 = indicators(A, 't3', 20, 0.7);
??? Error using ==> plus
Matrix dimensions must agree.

Error in ==> indicators at 689
t3 = c1*ema6+c2*ema5+c3*ema4+c4*ema3;

Do you know what the problem is? I'm knew to Matlab so having a hard time trying to decipher the code. SMA's and other stuff work fine. Cheers for the hard work.

[Nate Jensen]

Sorry for the delay, but I have finally added the True Strength Index. Work has been crazy busy and this unfortunately was moved down on the priority list.

I would like to point out that I have included 2 methods for calculating the ema in the tsi. The first method is very similar to Mike's, and is based off of Wikipedia's interpretation of the ema. The second method is based off of Matlab's interpretation of the ema. To choose one or the other, simply comment whichever method you do not wish to use. The default method is based off of Wikipedia.

[Nate Jensen]

Thanks again Mike!

I will take at look at your code this weekend and hopefully have something put together by Sunday.

[Mike]

Nate,
I had some time so I coded the True Strength Index (TSI). Probably not the most efficient code, and I don't know if it handles the NaN's properly, but basically it works with your program. I quickly checked the output and it looks right. Here it is:

% tsi - True Strength Index
% tsi = indicators(cl ,'tsi' ,fast,slow)

case 'tsi' % True Strength Index
% Input Data
close = vin(:,1);

% Variable Argument Input
if isempty(varargin)
fast = 3;
slow = 14;
else
fast = varargin{1};
slow = varargin{2};
end

momentum = [0; (close(2:end,1)) - close(1:end-1,1)];

slowEma = 0;
fastAbsEma = 0;
fastEma = 0;
slowAbsEma = 0;
vout = 0;

for i = uint16([2:size(momentum,1)])
slowEma(i,1) = (momentum(i,1) .* (2.0 ./ (1 + slow)) + (1 - (2.0 ./ (1 + slow))) .* slowEma(i-1,1));
fastEma(i,1) = (slowEma(i,1) .* (2.0 ./ (1 + fast)) + (1 - (2.0 ./ (1 + fast))) .* fastEma(i-1,1));
slowAbsEma(i,1) = (abs(momentum(i,1)) .* (2.0 ./ (1 + slow)) + (1 - (2.0 ./ (1 + slow))) .* slowAbsEma(i-1,1));
fastAbsEma(i,1) = (slowAbsEma(i,1) .* (2.0 ./ (1 + fast)) + (1 - (2.0 ./ (1 + fast))) .* fastAbsEma(i-1,1));

if(fastAbsEma(i,1) == 0)
vout(i,1) = 0;
else
vout(i,1) = 100 .* fastEma(i,1) ./ fastAbsEma(i,1);
end
end

[Nate Jensen]

Thanks Mike!

Yeah I could probably put that in.

[Mike]

Oh, I see you added these:
mfi - Money Flow Index
atr - Average True Range

excellent! Thank you.

Any plans to code the True Strength Index (TSI)?

[Mike]

 

[Nate Jensen]

The problem is fixed, and the new file should appear shortly.

I decided to use Matlab's datvecmx function to calculate the year, month, and day rather than try to figure it out on my own. I originally thought that I could write a script to calculate dates faster than Matlab's built-in functions could, but at the moment I can't even calculate those values correctly, much less faster than Matlab.

[Nate Jensen]

I forgot about this problem. The problem arises because I don't evaluate leap years correctly. I'll try to fix it in the next day or two.

[Bruno Pajusco]

I've first got the error by running the function using S&P exchange prices on the last 4000 observations. I have also tried with smaller subsets of other market data and even used my fake market data constructed from random variables but the error was still there. You could try to run it with your own data as I suspect you will get that same error but if you want me to send you the actual values for dates, open, high, low, close prices I would need your email address.

Bruno

[Nate Jensen]

Bruno,

Could you give me an example of how you are getting this error? You could email me your input that triggers the error, and I will see if I can reproduce it.

Mike,

Thanks for the positive feedback!

[Mike]

Great code, easy to use, and Very well documented! Thank you! Saved me a lot of time by not having to code this from scratch.

Note: When running this on Matlab version 2008 or older, the tilde (~) output argument is not supported. Simply change this to a dummy value (such as "val"), and the error will go away. Search for "~" and change lines: 407, 408, 802, 811, 858, 867. http://www.mathworks.com/help/techdoc/matlab_prog/bresuxt-1.html#br67dkp-1

[Bruno Pajusco]

Nate, I have come across a problem with the file while trying to estimate pivot points. The problem seems to be related to the code where you work out years, months and days from Matlab serial dates. In particular I had issues with the following block of code:

% Convert Matlab time to years, months, and days
temp_var1 = floor(date); % number of full days
temp_var2 = temp_var1/365.242199; % number of years
year = floor(temp_var2); % number of full years
if (4*(year/4-floor(year/4)) == 0) & ... % if it is a leap year
(100*(year/100-floor(year/100)) ~= 0) ...
| (400*(year/400-floor(year/400)) == 0) %#ok<AND2,OR2>
temp_var3 = floor((temp_var2-year)*366);% number of full days in year
month = cum_days1(temp_var3); % determine month
day = temp_var3-cum_days3(month); % determine day
else % if it's a normal year
temp_var3 = floor((temp_var2-year)*365);% number of full days in year
month = cum_days2(temp_var3); % determine month
day = temp_var3-cum_days4(month); % determine day
end

The code throws the following exception when constructing the variable month:

"Subscript indices must either be real positive integers or logicals.

Error in indicators (line 967)
month = cum_days2(temp_var3); % determine month"

I'm not sure why this is the case but when I construct similar variable to temp_var3 cum_days2 and then run that block of code it seems a lot happier than when I run the function. Any ideas?

Bruno

[Nate Jensen]

Krzysztof,

I suggest you simply write a new ema code that includes the previous ema value and you should be set. The following should do what you want,

function vout = ema(vin,prev_ema)

k = 2/(lag+1);
kvin = vin * k;
oneK = 1-k;

vout = kvin + (prev_ema * oneK);

end

[Krzysztof Fajst]

Hi,

I made some more investigation and I think tsmovavg based solution has a major drawback i.e. will not work for real time application. Lets assume you have 10k bars and you calculate ema on it. Than a new bar comes and calculate again 10k would be very time costly - much easier to calculate just last bar.

But in this case this code will not work because instead of use x-1 data
for initial value (so last data point of 1st calculation of ema) it will use some average values.

in other words it means that ema calculated on 10k+1 bars in one pass will have different value than ema calculated on last bar and it should not be a case.

Filter solution seems to be ok

Krzysztof

[Nate Jensen]

Krzysztof,

The method that I used is the method that Matlab uses to calculate the ema in their own moving average function 'tsmovavg'. This is not to say that that method is correct, but that is why I used it.

I briefly compared it to Google's ema on their finance page and the results were the same.

I would suggest to stick with what I have.

Nate

[Krzysztof Fajst]

It looks that there are at least 2 versions of code for EMA around i.e.
EMA calculated with algorithm from this function and EMA calculated with this code

% convert the period to an exponential percentage
ep = 2/(period+1);

% calculate the EMA
out = filter(ep,[1-(1-ep)],data,data(1)*(1-ep));

Unfortunately they give different results. Which one is correct ??

Krzysztof