3
$\begingroup$

At this moment, i need to do the analysis of rouble/us dollars exchange rate and the stock market index in Russia, I prefer to do that in a multivariate GARCH model. However, I have a question about the data. Currently, I have both data sets but they are not parallel, in other words, I have the stock market index in 30th Dec, 2014, 5th Jan, 2015, .... which is different from the exchange rate 31th Dec, 2014, 1th Jan, 13th, Jan, ..... I am not sure whether i downloaded the wrong data, otherwise, how should I deal with such case, to delete some days from both datasets?

$\endgroup$
  • $\begingroup$ Are you saying that both are sampled weekly but one on Mondays but the other on Wednesdays? $\endgroup$ – Kiwiakos Apr 15 '15 at 15:44
  • 1
    $\begingroup$ No, @Kiwiakos, I mean in different market, for example, A and B, one day 1, A is open but B is closed. $\endgroup$ – Fly_back Apr 15 '15 at 19:06
0
$\begingroup$

Usually I do one of the following:

  • I change granularity of data: download weekly returns instead of daily;
  • I create "super-laborious days index": I.e. (1) produce dates index consisting of all possible dates except for Sunday's and Saturday's; (2) on ALL series in dataset interpolate linearly dates, missing vs. super-index (say, on index A I have returns on A1, A2 and A7 dates, A3 were a national holiday; on B I have B1, B3 and B7: B2 were a holiday; 5 and 6 in both series are Sunday and Saturday; I create superindex C, with dates C1, C2, C3, C4 and then interpolate A3 between A2 and A4 and B2 between B1 and B3). I did that when wrote a paper about daily volatility spillovers in 12 countries across the world - many points where dates were unparalleled, as every nation has its own set of holidays.
$\endgroup$
  • 1
    $\begingroup$ Can you add a link to the paper you wrote? $\endgroup$ – Bob Jansen Sep 11 '15 at 10:58
0
$\begingroup$

Where there are missing days in a time series, any time series, I would prefer interpolation by some method to omitting the data when you are interested in a particular granularity. Interpolating data points can be done by a number of methods depending on how you look at the world and the best fit for your data:

  1. Assume that the previous period's close was the same as the data for the missing period (a "flat" model)
  2. Linearly interpolate the data, i.e. create a straight line between previous close and next open and set the data points off the line.
  3. Fit a spline or moving average through the data n periods on either side and interpolate off the curve.
  4. fit a more complicated model to the data n periods on either side and interpolate off that model.

realistically I normally choose 2 or 3 as the best payoff to effort as they best explain the "trend" over the closed period and out of hours / auction trading.

Another answer suggests changing the granularity which I think is undesirable if you are interested in interpolating or extrapolating on a daily basis rather than based on a higher granularity as you smooth out the granular characteristics.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.