I was hoping someone could recommend a good book on statistical methods for time-series on small data samples. We are really looking to do some forecasting/extrapolation given 13 years of previous data. I am looking at human resources data, and trying to understand variations in the gender balance in an organization. Hence we have information on men and women hired, promoted, and leaving from an organization over these 13 years. The observations are yearly, so if my metric is the percentage of female employees, then there are only 13 overall yearly observations--though we know these metrics at each organization level.

I know most time-series statistical methods don't make many assumptions about the data-generating mechanism. That feature of time-series models means that they often need a lot of data to give good estimates.

We are actually using a simulation based approach, since we know the data generating mechanism very well--in terms of hiring, attrition, and promotion practices.

Hence if anyone has any recommendations on time series forecasting based on small samples, please provide your suggestions or recommendations below. Thanks.


closed as unclear what you're asking by Nick Cox, Michael R. Chernick, Siong Thye Goh, Peter Flom - Reinstate Monica Oct 5 at 12:27

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • $\begingroup$ The title and the opening sentence clearly imply a book recommendation request, but then the question morphs into something else. Better to edit your question to have a specific and clear focus. $\endgroup$ – Nick Cox Oct 4 at 14:15

It is often possible to use small amounts of data to identify a useful ARIMA model. It all depends on the ratio of signal to noise that is in the data... the greater the ratio the fewer observations are needed to identify the model.

It is also important to possibly include "data cleansing" to deal with anomalies/pulses that otherwise might obfuscate the model identification phase.

If you post one of your "small time series" , I might be better able to help ... OR simply simulate a model and post the small data set .

I don't know of a book that particularly addresses this issue ....


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