Hi all,
Either for a cross-section or for a time-series of output observations, how to obtain potential levels of output so that one can estimate the output gap? Are there some simple or more logical methodologies in this front?
S.J. -
It sounds like you are talking about regression. Filling in those missing numbers is what is meant by "prediction," which may not have anything to do with forecasting. If you are talking about time series, you can still fill in for a missing number, and I think that is at least sometimes referred to as "backcasting." I made a great deal of use of "prediction" for missing data in finite population official statistics. The idea in all of these cases is to have a regression line (or curve) which gives you a relationship between the dependent variable (y), and one or more independent variables (x1, etc.), and for any missing y, you use the "predicted-y" from the regression line/ curve. For time series without regression, I suppose you can still interpolate and maybe extrapolate a little, but even this seems to have some similarity to regression, with time as x, and regression analysis seems the key to your question, overall.
But the regression you need may be tricky to decide upon. Even for simple regression through the origin, people often use ordinary least squares (OLS) without considering the natural heteroscedasticity that is there. Weighted least squares (WLS) should be considered then. Heteroscedasticity is not some "problem" to be "solved," but a natural part of the error structure. For multiple regression there are interactions between regressors to consider, as well as choice of variables, and nonlinear formats. (Note that leaving out important independent variables/regressors can lead to "omitted variable bias," and including unnecessary ones which may be too collinear increases variance.) There are regressions appropriate to different kinds of data (consider continuous versus count data). So regression is a large area to cover. Often graphics, such as scatterplots, can be helpful.
Your question seems to be quickly answered then by saying "use regression," but that covers a huge amount of material. Many books and articles fall under that broad topic.
Happy Holidays to All. - Jim
PS - I hope that I understood your question correctly. Perhaps this is not what you meant?
Potential output and output gaps are extensively used in the economic policy
making process. Potential output and output gaps are also
important concepts in economic research aimed at improving our understanding of how economies work.
Potential output can not be observed and has to be estimated using available (macro)economic data. A large number of methods have been developed for this purpose.
An overview and critical assessment of these methods is given in an OECD study published as:
Cotis, J.-P., Elmeskov, J., Mourougane, A. (2004) Estimates of Potential Output: Benefits and Pitfalls from a Policy Perspective, pages 35-53 in : L. Reichlin (ed) The Euro Area Business Cycle: Stylized Facts and Measurement Issues, London UK: Centre for Economic Policy Research.
Article ‘Estimates of potential output: benefits and pitfalls from a...
A simple and practical shortcut for manufacturing industry is to take differences of current level from the past highest ouput level. This assumes that at least once in the past the industry or firm was working at full capacity. The potential error is that if investment or other technological/organisation change have increased capacity, the measure will be biased downwards. If you have investment data and an estimate of their contribution to production, you can then correct the bias. On this line, you can find out other simple corrections for possible bias. Another approach that one can use is simply to look at inventory levels, and when they are at their minimum, you may assume the plant is working at full capacity, and again compute the output gap as a difference from such level.
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series