Skip to content

Uneven time series

4 messages · Luis Damiano, Michael Chen, Ilya Kipnis +1 more

Original
Luis Damiano
# 
Dear all,

I'm dealing with an uneven time series with regressors [0]. If the series
was regular, my preferred approach would be some form of space state
representation such as smoothing, time varying regression or an
autoregressive process with external regressors.

The series shows a clear trend and I believe that irregularity is an
important feature. So far, I've found two alternatives: a) continuous time
models and packages like ctsem [1] and cts [2], but they seemed limited in
their application, and b) exponential smoothing for uneven time series [3]
but they wouldn't allow for regressors.

Before I give up, I was wondering if you can suggest a better way to model
this series.

Thank you a lot!

Regards,

[0] http://puu.sh/s6O9q/ba6aa206f6.png

[1] https://cran.r-project.org/web/packages/ctsem/

[2] https://cran.r-project.org/web/packages/cts/

[3] https://oroboro.com/irregular-ema/
Michael Chen
# 
I have found in running my own customized strategy using blotter and updatePortf that whether I call updatePortf after completely applying my strategy/transactions or call updatePortf at regular intervals like weekly or monthly after each period's transactions, the performance statistics are the same.  So, is there a reason that I should updatePortf regularly during a run or just wait till the end?  I am puzzled.  

Thanks,

Michael
Ilya Kipnis
# 
Michael,

Unless you want to use a new end equity over the course of your strategy
(EG I started off with 100 dollars, I made 50 more dollars, I want to go
forward with 150 now), there's no reason to do so that I can think of.

-Ilya
On Sat, Nov 5, 2016 at 11:51 AM, Michael Chen <mwc4120 at hotmail.com> wrote:

            

      
      
      
    

        

      
      
    

        

  
  


  


      

    

    

      
      

        


  

        

      Joshua Ulrich
    

    

      
      #
    

  


  

      
On Sat, Nov 5, 2016 at 10:51 AM, Michael Chen <mwc4120 at hotmail.com> wrote:

            

      
      
      
    

        
Calling updatePortf marks your portfolio to market.  I assume (since
you didn't specify in your message) that your transaction sizing is
invariant to portfolio equity.  In that case, you should not expect
calling updatePortf to change strategy performance.

In short, if your strategy doesn't depend on your portfolio market
value, there's no need to call updatePortf until the end of the
simulation.