Skip to content

Align 5 minute bars

6 messages · Jeff Ryan, Noah Silverman, Brian G. Peterson

Noah Silverman
# 
And another question...

When I use the xts function to.minutes5(), I get a nice OHLC summary, BUT the time stamps are "messy".  Since my first and/or last ticks are not exactly on the minute, every 5 minute bar is now on some mid-minute frequency. 

Example:

to.minutes5(ds, indexAt="endof")
                    ds.Open ds.High ds.Low ds.Close ds.Volume
2007-01-01 18:34:44  882.50  883.75 880.50   880.50        83
2007-01-01 18:39:46  880.75  881.00 880.00   880.50        18
2007-01-01 18:44:52  880.25  880.25 879.50   880.00        25
2007-01-01 18:49:03  880.25  881.75 880.25   881.50        83
2007-01-01 18:52:11  881.50  881.50 881.50   881.50         9



Ideally, I'd love to have my bars aligned with round minute numbers.  i.e.
18:30
18:35
18:40
etc...

Any suggestions?

--
Noah Silverman
UCLA Department of Statistics
8117 Math Sciences Building
Los Angeles, CA 90095
Jeff Ryan
# 
Look at align.time. Bars are stamped to the last obs in the period in to.period. 

HTH
Jeff

Jeffrey Ryan    |    Founder    |    jeffrey.ryan at lemnica.com

www.lemnica.com
On Jun 16, 2011, at 2:25 PM, Noah Silverman <noahsilverman at ucla.edu> wrote:

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Noah Silverman
    

    

      
      #
    

  


  

      
Jeff,

If I understand the documentation correctly, align.time just shifts the time stamps of the existing bars.  That sounds dangerous as the individual ticks aren't moving.

For example, if a bar is currently set at:   18:39.46
We have a transaction at 18:39:48
As it stands, that transaction is in the *next* bar that would start at 18:39.47
If I use align.time, it will shift the bar to 18:40 BUT since we already have the OHLC summary, what happens to that transaction?


--
Noah Silverman
UCLA Department of Statistics
8117 Math Sciences Building
Los Angeles, CA 90095

        
On Jun 16, 2011, at 12:40 PM, Jeff Ryan wrote:

        

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Brian G. Peterson
    

    

      
      #
    

  


  

      
Your trade would have been *in* the bar.

typically, you'd use something of the type:

align.time(to.period(x,...)...) 

to first aggregate your trades into bars, and *then* align the bars.
This doesn't create any look-ahead bias.

  - Brian

        
On Thu, 2011-06-16 at 13:29 -0700, Noah Silverman wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Noah Silverman
    

    

      
      #
    

  


  

      
Good suggestion.  Exactly what I want.

However, in another message you suggested using aggregate or apply to generate my summary statistics.  I would need to do that before using to.period.  But, I want the summary states per-period.  (For example - number of transactions, number of up-ticks, etc.) 

The only way I can see to do this is the "long way".

1) Determine the endpoints of each bar, but don't convert to.period yet
2) Loop through each bar and generate the statistics from the transactions in that bar
3) Create a new data structure to hold the bar summaries

There must be a cleaner way.  Ideas?

--
Noah Silverman
UCLA Department of Statistics
8117 Math Sciences Building
Los Angeles, CA 90095

        
On Jun 16, 2011, at 4:19 PM, Brian G. Peterson wrote:

        

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Brian G. Peterson
    

    

      
      #
    

  


  

      
On Thu, 2011-06-16 at 19:54 -0700, Noah Silverman wrote:

            

      
      
      
    

        
Sort of the 'long way', I guess.

            

      
      
      
    

        
Here's an approach you might consider a little better.

1. run to.period on the trade prices to create your bars, 
2. assign the bars to a new variable
3. on the original data, period.apply() to generate each column of
summary stats
4. cbind each of these to the bars

You could do this all in one line, but that seems a little silly...