Data representation with string header

Hello,

I have a timetable and I want to know the correlation between diffrent variables in an organized way. For example, I have the closing price of Microsoft , Apple , Wallmart, and Cake. how can I join them and create a coverience matrix with their names on top and left of the matrix?

Answer 1

[Adam Danz]

If you have the Econometrics Toolbox you can use corrplot() to show a grid of correlations between all pairs of columns in your table - you can remove the columns that should be ignored.

If you don't have access to that function, you can use plotmatrix() and then add in the correlation values to make it look a lot like the corrplot by following this demo.

[Maha Almubarak]

thanks for your reply Adam. Do you know any function that would work on tiemetable ?

[Adam Danz]

You can use either of these option after eliminating data that cannot be converted to double precision.

For example,

% load built-in timetable example
load outdoors

% isolate data and eliminate the datetime values
corrplot(outdoors.Variables)

Or if you want to isolate specific columns,

corrplot([outdoors.Humidity, outdoors.TemperatureF])
% or
corrplot(outdoors{:,[1,2]})

[Maha Almubarak]

I did this corrplot([NSJN.Close_SP,NSJN.Close_NIKKEI,NSJN.JPY]);

It gave me var 1 var 2 and so on it did not give me my actual variable names.

[Adam Danz]

You've got two options.

1) corrplot can receive a table as an input and it will use the var names from the table. You'll have to convert yoru timetable to a table using timetable2table and then isolate the desired columns so the output is still a table using indexing T(:,[2,3]), for example.

2) Use the matrix input that you're already using and specify the var names using the 'varNames' property which you could either manually enter or pull from the timetable using T.Properties.VariableNames where T is your timetable or table.

Important note: var1 and var3 clearly do no obey the linearity assumption so their correlation values are meaningless. The other pairs of variables do not obey the normality assumption nor do they appear to be on a continuous scale. I don't see a single scatter plot that looks like it would produce a meaningful correlation value.

[Maha Almubarak]

thanks Adam for the reply. Appreiat the time :)