Estimate coefficients to predict out of sample for the excluded quarter.

[Sydul Karim]

Hi, I want to evaluate the model’s (reg lnUnmprate lnInflation lnFedfunds) ability to predict out of sample. Specifically, I want to start by removing the first quarter(qrt204) from the estimation sample and use the estimated coefficients to predict out of sample for the excluded quarter. I then want to repeat this for each quarter and report the RMSE over all quarters for the model specification to compare it with a benchmark model.(not mentioned here). Sample time series data is given below;

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(qrt Inflation Unmprate Fedfunds)
204   .9101716 2.2009215  -1.853635
205  1.1963816 2.2046046  -2.371578
206    .977927 2.1972246  -2.484907
207  .07792359 2.1556306   -2.61274
208  1.0651164 2.1122313 -2.2697952
209   .7287415  2.104134  -1.875141
210  1.0592687 2.0835996 -1.9425824
211   .7660922 2.0541236 -1.8325815
212   .6979457   2.04554 -1.9425824
213  .50716305 2.0193377 -2.1484344
214   .9642801    1.9787  -2.484907
215   .9775378 1.9363407  -2.445686
216    .654176   1.89712   -2.61274
217   1.055461 1.8245493  -2.371578
218   .8935435 1.8028094 -2.4079456
219  -.1565022  1.740466 -2.3025851
220  -1.490751 1.7107904  -2.207275
221  1.0335808 1.6925528 -2.0928645
222    .497876 1.6292405 -1.9902104
223  -.5362221 1.6160824 -1.8325815
224 -1.4332616  1.596015 -1.0216513
225  1.1255546 1.5892352  -.9852836
226    .614585 1.5892352   -.924659
227   .9833136 1.5616473  -.7985077
228   .8862488 1.5260563  -.3566749
229   .4922492     1.474  -.0512933
230   .9551601  1.458615   .1426563
231  1.0736006 1.4190842   .1850955
232   .8870767 1.4028237   .3692621
233   1.237515 1.3609766  .55196756
234   .8084829  1.335001   .6540598
235   .7520165  1.335001   .7975072
end
format %tq qrt

Any help will be much appreciated.

Regards,
Karim

[Phil Bromiley]

I guess I'd look over quarters and run something like:

g sumpred=.

forvalues q=204/235 {
reg y x if qrt !=`q'
predict yhat`q' if qrt==`q'
replace sumpred=yhat`q' if qrt==`q'
}

I haven't checked whether this runs, but you should get the idea.

[Sydul Karim]

Thanks, Phil. I think this is what I wanted.