Skip to content

Missing data?

13 messages · michael.weylandt at gmail.com (R. Michael Weylandt, Kevin Burton, R. Michael Weylandt +1 more

michael.weylandt at gmail.com (R. Michael Weylandt
# 
Couldn't you use seq.Date() to set up the time index and then just fill as appropriate?

Alternatively, to.weekly if you are starting with a daily series. 

Michael
On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <rkevinburton at charter.net> wrote:

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Kevin Burton
    

    

      
      #
    

  


  

      
Thank you for the suggestions.

The only problems I see with 'to.weekly' is converting from the OHLC format
and realizing that the date is the last day of the week rather than the
first day of the week. Very minor compared to doing the whole thing myself.

-----Original Message-----
From: R. Michael Weylandt <michael.weylandt at gmail.com>
[mailto:michael.weylandt at gmail.com] 
Sent: Tuesday, November 22, 2011 3:10 PM
To: Kevin Burton
Cc: <r-help at r-project.org>
Subject: Re: [R] Missing data?

Couldn't you use seq.Date() to set up the time index and then just fill as
appropriate?

Alternatively, to.weekly if you are starting with a daily series. 

Michael

On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <rkevinburton at charter.net>
wrote:

            

      
      
      
    

        
series.

            

      
      
      
    

        
end date.

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Kevin Burton
    

    

      
      #
    

  


  

      
Also with to.weekly there seems to be a problem with when the week starts.
For example:

            

      
      
      
    

        
as.Date("2011-10-09"), as.Date("2011-10-10")), frequency=52)

            

      
      
      
    

        
xts.ts.Open xts.ts.High xts.ts.Low xts.ts.Close
2011-01-01           1           1          1            1
2011-01-10           2           2          2            2
2011-10-09           3           3          3            3
2011-10-10           4           4          4            4

            

      
      
      
    

        
as.Date("2011-10-09"), as.Date("2011-10-10")), frequency=52)

            

      
      
      
    

        
xts.ts.Open xts.ts.High xts.ts.Low xts.ts.Close
2011-01-02           1           2          1            2
2011-10-09           3           3          3            3
2011-10-10           4           4          4            4

So in the first case the week ends on January 1st. But the second indicates
that the end of the week is the 2nd but it includes the data from the first.
I would expect that the first column should be consistent.
Notice that 10-09 and 10-10 are properly considered different weeks because
the 9th is a Sunday and the 10th is a Monday (the beginning of the week).

-----Original Message-----
From: R. Michael Weylandt <michael.weylandt at gmail.com>
[mailto:michael.weylandt at gmail.com] 
Sent: Tuesday, November 22, 2011 3:10 PM
To: Kevin Burton
Cc: <r-help at r-project.org>
Subject: Re: [R] Missing data?

Couldn't you use seq.Date() to set up the time index and then just fill as
appropriate?

Alternatively, to.weekly if you are starting with a daily series. 

Michael

On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <rkevinburton at charter.net>
wrote:

            

      
      
      
    

        
series.

            

      
      
      
    

        
end date.

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Kevin Burton
    

    

      
      #
    

  


  

      
Void of any other suggestions this approach makes sense but for my case I
think I need to use zoo objects rather than xts. If I sequence the data
generally I don't know if there will be 365 days in the year or 366. So I
have to sequence the dates as:

seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day")

If I use this sequence with xts I get:

            

      
      
      
    

        
by="day"))
Error in xts(NA, seq(from = as.Date("2011-01-01"), to =
as.Date("2011-12-31"),  : 
  NROW(x) must match length(order.by)

If I leave the 'data' empty I don't get the error but if I try to assign an
individual item (fill as appropriate)

            

      
      
      
    

        
by="day"))

            

      
      
      
    

        
Error in structure(coredata(x), names = x.attr$dimnames[[1]]) : 
  'names' attribute [365] must be the same length as the vector [358]

So now I need to remember that I have not filled in all of the data. Also
simple dereferencing gives:

            

      
      
      
    

        
Error in `[.xts`(ds, 1) : subscript out of bounds

With zoo I am able to create a time-series where all of the data is
initially NA:

            

      
      
      
    

        
by="day"))

So I can fill the data as appropriate and the remaining slots will have NA.
I may be new with xts but I cannot see a way of creating a useable 'blank'
time-series.

Also with xts it seems like the frequency is ignored.

            

      
      
      
    

        
by="day"), frequency=52)

            

      
      
      
    

        
[1] 1

Whereas zoo remembers the frequency setting

            

      
      
      
    

        
by="day"), frequency=52)

            

      
      
      
    

        
[1] 52

But since the ultimate goal is to get the time-series in a 'ts' format (as
many functions require 'ts') it seems like even zoo has problems:

            

      
      
      
    

        
Time Series:
Start = c(14975, 1) 
End = c(15339, 1) 
Frequency = 52 
    [1]   1  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA
NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA
NA  NA  NA  NA  NA
   [42]  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA   2  NA  NA  NA  NA  NA
NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA
NA  NA  NA  NA  NA
   [83]  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA
NA  NA  NA  NA  NA   3  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA
NA  NA  NA  NA  NA
  [124]  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA
NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA   4  NA  NA
NA  NA  NA  NA  NA
  [165]  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA
NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA  NA
NA  NA  NA  NA  NA
  [206] . . . . . .
 So the conversion from zoo to ts maintained the frequency but I am not sure
where it decided on the start and end values. Also the conversion seemed to
changed the data also. Notice that every period (52 entries) the original
data is maintained. In other words if ds is the original zoo time series
then ds[1] is 1 and ds[2] is 2 etc. The converted time-series keeps ds[1]
but inserts 51 NA's then adds ds[2] etc till the end of the series.  That is
not what the initial data was. The conversion is inserting data of its own.

The conversion to ts from xts seems better behaved:

ds <- xts(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"),
by="day"), frequency=52)

            

      
      
      
    

        
Time Series:
Start = 1 
End = 365 
Frequency = 1 
  [1]   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15  16  17
18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36
37  38  39  40  41  42
 [43]  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59
60  61  62  63  64  65  66  67  68  69  70  71  72  73  74  75  76  77  78
79  80  81  82  83  84
 [85]  85  86  87  88  89  90  91  92  93  94  95  96  97  98  99 100 101
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120
121 122 123 124 125 126
[127] 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143
144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
163 164 165 166 167 168
[169] 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204
205 206 207 208 209 210
[211] 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227
228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246
247 248 249 250 251 252
[253] 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269
270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288
289 290 291 292 293 294
[295] 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311
312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
331 332 333 334 335 336
[337] 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353
354 355 356 357 358 359 360 361 362 363 364 365

But alas the frequency is ignored.

So this is what I have found out using these two packages.  If I want to
create a 'blank' data set it seems like zoo is 'better' since I can create a
time-series initialized with NA irrespective of the length of the series.
However I must be unfamiliar with the conversion because zoo doesn't convert
to a regular 'ts' very well.  But  zoo remembers the frequency setting
whereas xts just ignores it.

It seems like there is still considerable work to solve the original
problem. If I create a time series and fill in the values that are
appropriate I still could have NA in the series it seems to.weekly has a
problem with NA in the time series:

            

      
      
      
    

        
to=as.Date("2011-12-31"), by="day"), frequency=52)

            

      
      
      
    

        
Error in if (drop.time) x <- .drop.time(x) : 
  argument is not interpretable as logical
In addition: Warning message:
In to.period(x, "weeks", name = name, ...) :
  missing values removed from data


-----Original Message-----
From: R. Michael Weylandt <michael.weylandt at gmail.com>
[mailto:michael.weylandt at gmail.com] 
Sent: Tuesday, November 22, 2011 3:10 PM
To: Kevin Burton
Cc: <r-help at r-project.org>
Subject: Re: [R] Missing data?

Couldn't you use seq.Date() to set up the time index and then just fill as
appropriate?

Alternatively, to.weekly if you are starting with a daily series. 

Michael

On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <rkevinburton at charter.net>
wrote:

            

      
      
      
    

        
series.

            

      
      
      
    

        
end date.

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    
3 days later

    

      
      

        


  

        

      R. Michael Weylandt
    

    

      
      #
    

  


  

      
Why do you need to use a frequency attribute for these data? The point
of the zoo/xts line of time series implementations is that the time
stamps are carried through for each observation (unlike ts) and can be
irregular. Both classes exist precisely to avoid being forced into a
frequency attribute.

As far as setting up the time elements, wouldn't this work? Change the
start date to get weeks on any desired day

d <- seq.Date(from = as.Date("2011-11-26"), by = -7, length.out = 100)
xts(rep(NA, length(d)), d)

You can avoid the OHLC formatting of to.weekly if you want with the
OHLC = FALSE parameter. And if you want to index it by the first of
the week rather htan the last, just try this:

time(x) <- time(x) - 6

Michael

        
On Tue, Nov 22, 2011 at 6:50 PM, Kevin Burton <rkevinburton at charter.net> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Gabor Grothendieck
    

    

      
      #
    

  


  

      
On Tue, Nov 22, 2011 at 6:50 PM, Kevin Burton <rkevinburton at charter.net> wrote:

            

      
      
      
    

        
The problem is that you seem to want a fixed number of periods per
year but there is not a constant of 52 weeks nor 365 days in a year.
You are going to have give up something since your apparent criteria
conflict with reality.  For example, you could use months in which
case there are exactly 12 or you could stick more than 7 days into the
first or last week of the year so that there are exactly 52 weeks in a
year but they don't all have the same number of days, etc.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Kevin Burton
    

    

      
      #
    

  


  

      
I admit it isnt reality but I was hoping through judicious use of these functions I could approximate reality. For example in the years where there are more than 53 weeks in a year I would be happy if there were a way to recognize this and drop the last week of data. If there were less than 53 I would "pad" the year with an extra dummy week. This is just about the same as your suggestion of putting more than 7 days in the first and last weeks. But i still need this kind of date manipulation to even know how many days to add in to make the approximation viable. This kind of best approximation to reality seems better than to settle for the resolution of a month just because it is consistent. Daily would be too much data and even then there would be an approximation due to leap years.

        
On Nov 26, 2011, at 3:13 PM, Gabor Grothendieck <ggrothendieck at gmail.com> wrote:

        

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Kevin Burton
    

    

      
      #
    

  


  

      
I was just trying to be complete. Why is the frequency argument and
attribute available?

-----Original Message-----
From: R. Michael Weylandt [mailto:michael.weylandt at gmail.com] 
Sent: Saturday, November 26, 2011 2:40 PM
To: Kevin Burton
Cc: r-help at r-project.org
Subject: Re: [R] Missing data?

Why do you need to use a frequency attribute for these data? The point of
the zoo/xts line of time series implementations is that the time stamps are
carried through for each observation (unlike ts) and can be irregular. Both
classes exist precisely to avoid being forced into a frequency attribute.

As far as setting up the time elements, wouldn't this work? Change the start
date to get weeks on any desired day

d <- seq.Date(from = as.Date("2011-11-26"), by = -7, length.out = 100)
xts(rep(NA, length(d)), d)

You can avoid the OHLC formatting of to.weekly if you want with the OHLC =
FALSE parameter. And if you want to index it by the first of the week rather
htan the last, just try this:

time(x) <- time(x) - 6

Michael

On Tue, Nov 22, 2011 at 6:50 PM, Kevin Burton <rkevinburton at charter.net>
wrote:

            

      
      
      
    

        
NA.

            

      
      
      
    

        
The conversion is inserting data of its own.

            

      
      
      
    

        
series.

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      R. Michael Weylandt
    

    

      
      #
    

  


  

      
Back compatibility with other time series best I can tell, but to be
honest, I'm not even sure how it plays into that. Perhaps it's just an
artifact in the signature.

It doesn't seem to have a role in the xts constructor. E.g.,

identical(xts(1:5, Sys.Date()+1:5, frequency = 1), xts(1:5,
Sys.Date()+1:5, frequency = 3))

Michael

        
On Sun, Nov 27, 2011 at 4:51 PM, Kevin Burton <rkevinburton at charter.net> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Gabor Grothendieck
    

    

      
      #
    

  


  

      
On Sun, Nov 27, 2011 at 4:08 PM, Kevin Burton <rkevinburton at charter.net> wrote:

            

      
      
      
    

        
OK. As you are willing to regard days past the 364th as part of the
last week of the year then we can do this.

Create a zoo object z as test data.   Then convert its time scale to
year + week/52 where 0 is the first week of the year and we replace
any week that is greater than 51 with 51.  Then we aggregate z by week
taking the last data point in the week and convert it to ts.  Because
of the way we constructed it the frequency will be 52.

library(zoo)

# test data
z <- zoo(1:100, Sys.Date() + 1:100)

yr.wk <- with(as.POSIXlt(time(z)), year + 1900 + pmin(yday %/% 7, 51) / 52)
z.wk <- aggregate(z, yr.wk, tail, 1)
z.ts <- as.ts(z.wk)

frequency(z.ts) # 52

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Kevin Burton
    

    

      
      #
    

  


  

      
This has been very helpful. Thank you.

At the risk of further confirming my ignorance and taxing your patience I
would like to add another question. How would I modify this code so that
each week starts with the same day of the week regardless of the year? I
would add this stipulation so that for multiple years I always get the same
'week-number' like

            

      
      
      
    

        
[1] "47-0"

The convention (at least for US culture) seems to be that the week starts
with Sunday (it is index 0 for day of week). So it would be convenient if
the code was modified so that each 'week' began on Sunday. The partial at
the beginning would just start with the day of week that was at the start. I
still would want to aggregate that 'week-number's that are greater than 51
like you have shown.

Thanks again.

Kevin
-----Original Message-----
From: Gabor Grothendieck [mailto:ggrothendieck at gmail.com] 
Sent: Sunday, November 27, 2011 4:24 PM
To: Kevin Burton
Cc: r-help at r-project.org
Subject: Re: [R] Missing data?

On Sun, Nov 27, 2011 at 4:08 PM, Kevin Burton <rkevinburton at charter.net>
wrote:

            

      
      
      
    

        
functions I could approximate reality. For example in the years where there
are more than 53 weeks in a year I would be happy if there were a way to
recognize this and drop the last week of data. If there were less than 53 I
would "pad" the year with an extra dummy week. This is just about the same
as your suggestion of putting more than 7 days in the first and last weeks.
But i still need this kind of date manipulation to even know how many days
to add in to make the approximation viable. This kind of best approximation
to reality seems better than to settle for the resolution of a month just
because it is consistent. Daily would be too much data and even then there
would be an approximation due to leap years.

            

      
      
      
    

        
OK. As you are willing to regard days past the 364th as part of the last
week of the year then we can do this.

Create a zoo object z as test data.   Then convert its time scale to
year + week/52 where 0 is the first week of the year and we replace any week
that is greater than 51 with 51.  Then we aggregate z by week taking the
last data point in the week and convert it to ts.  Because of the way we
constructed it the frequency will be 52.

library(zoo)

# test data
z <- zoo(1:100, Sys.Date() + 1:100)

yr.wk <- with(as.POSIXlt(time(z)), year + 1900 + pmin(yday %/% 7, 51) / 52)
z.wk <- aggregate(z, yr.wk, tail, 1) z.ts <- as.ts(z.wk)

frequency(z.ts) # 52

--
Statistics & Software Consulting
GKX Group, GKX Associates Inc.
tel: 1-877-GKX-GROUP
email: ggrothendieck at gmail.com

  
  


  


      

    

    

      
      

        


  

        

      Gabor Grothendieck
    

    

      
      #
    

  


  

      
On Sun, Nov 27, 2011 at 8:10 PM, Kevin Burton <rkevinburton at charter.net> wrote:

            

      
      
      
    

        
It would be the same except replace the calculation of yr.wk with:

tt <- time(z)
yr.wk <- as.numeric(format(tt, "%Y")) + pmin(as.numeric(format(tt,
"%W")), 51)/52

This puts the 1st Sunday of the year and the days prior to it in week
0, the next 7 days are in week 1 and so on.  All days after the 51st
Sunday are either in week 51 or are forced to be.