General Discussion - NM DEV Forum

How to calculate the Akaike Information Criterion (AIC)

Anonymous — Tue, 30 Jul 2024 16:51:25 +0000

In below code I return the "pValue" in the "AugmentedDickeyFuller".
Now we can use different LagOrder parameters where it is important to choose the "best" possible lagOrder to return as accurate "pValue" as possible.

Normally a good way to determine what LagOrder value to choose is by calculate the
Akaike Information Criterion (AIC).

Where the lowest found AIC would be a good pValue to choose.

I am not sure how to calculate the AIC with the library in this context?
Below line is just pseudo code and that function do not exist but show the intent.

Thank you!

CalculateAIC(statistic, lagOrder)

TrendType P_1 = TrendType.CONSTANT;
ADFDistribution dist = P_1.getDistribution(array_MSFT.Length);

double minAIC = double.MaxValue;
int bestLagOrder = 0;

int maxLagOrder = 20;
for (int lagOrder = 0; lagOrder <= maxLagOrder; lagOrder++) {
    AugmentedDickeyFuller ADFtest = new AugmentedDickeyFuller(array_MSFT, P_1, lagOrder, dist);
    double pValue = ADFtest.pValue();
    double statistic = ADFtest.statistics();

    //How to Implement AIC calculation? (Pseudo Code)
    double aic = CalculateAIC(statistic, lagOrder); 
    
    if (aic < minAIC) {
        minAIC = aic;
        bestLagOrder = lagOrder;
    }
}

How to declare ADFDistribution for AugmentedDickeyFuller in C# code

Anonymous — Mon, 29 Jul 2024 01:06:42 +0000

Hello,

I am trying to understand how to create the: AugmentedDickeyFuller

The parameter I have problem to declare is the below ADFDistribution

ADFDistribution ADF_dist

How to implement that correctly for this C# code?

Thank you!

            double[] array_MSFT = new double { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 };
            int lagOrder = 0;
            ADFDistribution ADF_dist = new ???
            AugmentedDickeyFuller ADFtest = new AugmentedDickeyFuller(array_MSFT, TrendType.CONSTANT, lagOrder, ADF_dist);

            double pValue = ADFtest.pValue();
            double statistic = ADFtest.statistics();
            MessageBox.Show(pValue + "__" + statistic);

Using the library in an open source project

Anonymous — Sun, 19 Feb 2023 12:12:16 +0000

I'm the main developer of an open source algo-trading platform and SuanShu or NMDev seems to be a good library to base some more advanced strategies upon.

As far as I can tell, only SuanShu (2012 release as available on GitHub) comes under the open source Apace license that would allow me to modify and redistribute it. Already great that this version has been open sourced 👍.

NM Dev om the other hand requires a commercial license by the end user. That would prohibit me from releasing my own software for free since if I include nm.dev library (even if I pay for it and only include the binary), I cannot redistribute it.

Any insights (especially if I'm wrong), would be very welcome? Thanks in advance for any help!!!

For those interested in the open source algo-trading project, checkout http://roboquant.org

Kind regards,

Peter

nmdev, the numerical method Java library, version 1.2.4 has been released

haksunli — Fri, 04 Nov 2022 10:39:13 +0000

nmdev version 1.2.4 has been released. This is a minor version update and small bug fixes.

Purpose of buying Numerical Methods using java....

citroenie — Thu, 15 Sep 2022 23:17:32 +0000

What't the purpose of buying the book 'Numerical Methods using java...' when the free version (https://github.com/nmltd/SuanShu/tree/master/20120606%20src/src/main/java/com/numericalmethod/suanshu) is a different librarie which isn't usable (chapter 5) or when the nmdev-1.0.0 download has a licence which is expired at 2020-05-12

Can you give a new trial-licence for nmdev-1.0.0

Kind regards

Arnoud

How to be sure that two series are cointegrated with CointegrationMLE and JohansenTest

Anonymous — Tue, 28 Jun 2022 08:58:39 +0000

Hello everyone!
I am working on my current java-based project where i have to do assumption is two series cointegrated or not. And i faced with problem that there are not enough duscussions and community-envolving about nm-dev library.
If i used Engle-Granger cointegration test, then i just calculated P-value, T-value, C-value. So i can apply only one condition to bu sure series are cointegrated if (P-value < 0.5 && T-value < C-value).
It is very easy to do with python library "statsmodels" just importing "coint" function that returns all three variables above.

So i need your helping with better understanding how Johanson test works and how can i apply it to my series to do exactly conclusion about cointegration.

So, i have initialized an instance of the class CointegrationMLE

CointegrationMLE cointegrationMLE = new CointegrationMLE(timeSeries, false);

And i have numerous parameters, such as

n: 198
rank: 2
eigen values: 
alpha: 2x2
 
 -0,087240, 0,001106,
 0,175470, 0,002975,
1-th cointegrating factor: 
2-th cointegrating factor: [1,000000, -0,051580

Are these parameters sufficient to make an assumption about series cointegration? And what is the final rule for it? Do I need to additionally use JohansenTest?