Pay special attention to vintage data when replicating a paper

This is written for my friends/students, so it should be in English. If you have an interest in replicating a paper, but you are unsure about what is written bellow, just drop me a message on the website or send me an email, I will be more than happy to discuss with you:
From my point of views, it is always interesting to replicate a paper because it: (i) makes you understand thoroughly the paper; (ii) improves your studying skills; and (iii), most importantly, is that if you can not replicate that paper, you can not definitely make a better paper than that paper. For the “better” word, I mean a more complicated model with new factors which could be considered as one of prerequisite conditions for a PhD thesis.
However, there is a common mistake/error when one is trying to replicate a paper that I call “data-lead mistake”. So, what is “data-lead mistake”?
Definition: I think that the best way to express a definition is to take an example. Assuming that, on a very nice day, say tomorrow: 12/11/2011- hopefully it will be nice-, you would like to replicate Taylor rule in the famous paper Taylor (1993), named “Discretion versus Policy Rules in Practice” which used data of the US economy (inflation rate, output, federal funds rate) from 1987q1-1992q4.
The first action often taken is to “look for data with the same period” from some well-known sources, especially IFS (International Financial Statistics) and then set up the range “1987q1-1992q4” and click “download” button. If you can find estimations highly similar with that of Taylor (1993), lucky you. However, it’s rare. In most cases, there will be some differences, even large ones, if you follow such a strategy.
This is definitely not your mistake, but the problem here is that: Taylor did that paper on the quarter 1, 1993, while you are going to do that on the quarter 4, 2011. For this, I mean that you are observing the data over the period 1987q1-1992q4 by available “updated” information of today, say on IFS accessed 12/11/2011, which is different from those observed by Taylor in 1993. For instance, for real output in the first quarter of 1986, in terms of Billions of real dollars and seasonally adjusted, Taylor saw it as 4460, while you find it equal 7186.9 (strictly speaking, 7186.9 is the data available from the quarter 3 2011) ; this is not because of a change in the base year – in fact, these figures are calculated on the same base year. You can find the reasons of such differences from the links below. In terms of jargon, data observed by Taylor is called the q1-1993 vintage, while the q3-2011 vintage is noted for your observations.
Therefore, to replicate the Taylor rule 1993, you should use the q1-1993 vintage. Otherwise, there may be a possibility of “data-lead mistake” in your replication.
-So, where can you find the vintage data?
Such a data for the US economy is available to download free-of-charge at Real-Time Data Research Center, Federal Reserve Bank of Philadelphia. Particularly,
+ For the historical Data Files for the Real-Time Data: http://www.philadelphiafed.org/research-and-data/real-time-center/real-time-data/data-files/
+ To see what central bankers in the past forecasted the inflation, real output growth, so on (For example, what central bank’s staff thought about inflation in the quarter 4, 2000 with the q4-1999 vintage): see Survey of Professional Forecasters or Greenbook Data Sets

Hope you find interesting points from this note. Good luck for your studying!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s