| in the sample, of course together with appropriate changes in certain parts of the model to allow it to accommodate the wartime government policies. 5) The use of quarterly data would multiply the effective sample size by approximately four,72 thus producing more accurate estimates, provided the problem of serial correlation can be solved (see next item). | |
| 6) The development of practical methods of estimation that do not require the assumption of zero serial correlation of disturbances would be useful. As already mentioned, Chernoff and Rubin have worked on this problem but as yet no attempt has been made to use their results. | |
| 7) Mathematical (or experimental)73 studies to determine the size of the small-sample bias in the estimation of structural parameters by the least- squares method and by the various maximum likelihood methods would be very helpful in deciding which procedure to use. | |
| 8) Studies might be made of the effect of estimating the parameters of a model by using data generated by a structure not belonging to the given model, i.e., the effect of estimating from the wrong model. This is a general problem which includes the case of estimating by the least squares method when to do so is not theoretically justified. If a "slightly incorrect" model always or often leads to absurd results, the type of econometrics presented in this paper will suffer a severe setback, because we know from the start that our models are at least slightly incorrect. | |
| 9) It would be interesting, though expensive, to estimate the parameters of a fairly large system of equations by the full information maximum likelihood method and analyze the results. But this would not be likely to be immediately useful in getting better estimates unless the sample size were much larger than 22. |
Appendix A TIME SERIES
Until 1942 all time series are as given in Klein (11), pp. 141-3, except that those marked with an asterisk below have been revised as indicated
____________________72
The effective sample size would be multiplied by exactly 4, except for several small
points: the fact that one degree of freedom goes into the estimation of each parameter; the possibility of adding four new parameters in order to allow for seasonal
changes (this is done by introducing four new exogenous variables x1, x2, x3, x4, such
that in the ith quarter all are 0 except xi, which is 1, and estimating the parameter of
each); etc.
73
Orcutt and Cochrane (18, 19) have used sampling experiments of a type that
might be widely applied in getting information of any desired degree of statistical
reliability about certain problems that seem to be secure against direct mathematical
attack.
-89-
Questia, a part of Gale, Cengage Learning. www.questia.com
Publication information:
Book title: Conference on Business Cycles.
Contributors: National Bureau Committee for Economic Research - OrganizationName.
Publisher: National Bureau of Economic Research.
Place of publication: New York.
Publication year: 1951.
Page number: 89.
This material is protected by copyright and, with the exception of fair use, may not be further copied, distributed or transmitted in any form or by any means.
One moment ...
- Georgia
- Arial
- Times New Roman
- Verdana
- Courier/monospaced
Reset