how can i extrapolate the last value of a matrix?

Dear all,

I have this matrix

 clear all

A={

   '02/2000'                  [NaN]        [NaN]  
    '03/2000'              [NaN]                [3] 
   '04/2000'               [NaN]               [5]
    '05/2000'              [4]                  [4]
   '06/2000'               [3.9313]             [0.9313]
    '07/2000'               [3.9313]             [3.9313]
    '08/2000'              [NaN]}                  [45]} 
    '09/2000'                  [4.1583]        [44.1583]
    '10/2000'              [3.9389]              [334.9389]
   '11/2000'               [4.3944]              [45.3944]
    '12/2000'              [4]                        [4.56]
   '01/2001'               [3.9313]                      [56.9313]
    '02/2001'               [3.743]                  [4.9543]
    '03/2001'              [NaN]                       [NaN]}

 my goal is to extrapolate the values of the last observations, that is,  the observations that corresponds to the
 the date   '03/2001' 

 Is there a way to do that in matlab?

 Note that I have many such vectors and  the last date for each vector is different.

 Cheers

Answer 1

[Jos (10584)]

You have to define a model first. As an example, assume you have data points (time, value) like this:

Time  = [1 3 6 7 8]
Value = [10 11 15 19 25]

and you want to get the value at an unknown time. You could, for instance, fit a linear model "Value = a x Time +b" through the known data points by which you get the parameters a and b, and in turn retrieve the predicted values at unknown times.

p = polyfit(Time,Values,1) 
Value10 = polyval(p,10)

However, there are many, many issues to be aware of! For instance, what is the underlying model? Linear or not? Also, extrapolation is often very tricky ...

[tzaloupas]

thanks Jos. I am interested in both polyfit() and extrapolation.

Could you please be a bit more specific about the code that I have to use in my question?

thanks

[José-Luis]

You have not answered Jos question. What sort of data is this? What sort of model are you using? To blindly extrapolate is a bad habit. Extrapolation is dangerous and can have disastrous consequences (the crash of the Challenger, e.g.)

[tzaloupas]

thanks Jose . My data are on prices of goods. I tried to extrapolate the last observation and in some cases I obtain negative prices

I do not have a specific model in my mind. Is there anything you can do to help me with the above example?

thanks

[José-Luis]

I am sorry, but if I had a reliable model to extrapolate the price of goods I would be a very very very very rich man. What you are asking is not simple, to say the least.

Answer 2

[Image Analyst]

If you want to fit, say, 123 locations between -2 and 42 you could create your x data:

xFitted = linspace(-2, 42, 123);

Now get the fit:

monthNumber = % Extract this from column 1
column2 = % Extract this from column 2.
coefficients = polyfit(monthNumber , column2 , 1); % Or whatever order of polynomial you want.

Then do the fit/extrapolation:

yFitted = polyval(coefficients, xFitted);

yFitted will be the fitted y values between -2 and 42. Since your data went from month 2 to month 15, values from -2 to 2 and from 15 to 42 are extrapolated. Values between 2 and 15 are fitted/interpolated/regressed because they occur within the valid training range of your data.

[tzaloupas]

thanks Image analyst. Still can not understand how to apply your appraoch to my example. could you help me a bit more?

thanks

[tzaloupas]

please? I am struggling to find the answer and I cant

[tzaloupas]

anybody else?