README.md

In jtelliott/attrition-forecast: Forecast US Air Force Officer Attrition Using Regression with ARIMA Errors

attrition-forecasting

1st Lt Jake Elliott 18 January 2018

Analyzing the Economic Effects on US Air Force Officer Attrition

Howdy! This package utilizes freely available economic data (from the St. Louis FRED) and personnel data (provided by HAF/A1) to forecast US Air Force officer monthly attrition rates. The package provides a reproducible means for personnel analysts (and any other interested parties) to analyze the effects of the economic environment on attrition.

The package allows the user to select from a host of economic indicators such as the U1 unemployment rate, labor market momentum index, labor force participation rate, etc. to generate a mathematical model for descriptive and predictive purposes. This endeavor is geared to those interested in investigating the relationship between aspects of civilian labor markets and attrition in the military. Additionally, the model(s) developed may be used to forecast attrition based on economic data.

The package aids the development of a regression model with ARIMA errors, and as such a basic understanding of statistical model evaluation and assessment is suggested. Additionally, a understanding of, or willingness to research, labor market inidicators will be useful in model specification. The package relies primarily on the forecasting functions offered by Rob Hyndman’s fpp2 R package. We build off of this work by focusing efforts on regression and ARIMA models, and their evaluation and specification.

More specifically, this package provides the following features:

• Selection of economic variables for regression from a pre-determined subset, included with the package (any additional indicators will have to be manually downloaded from the St. Louis FRED)

• Specification of the personnel data subset, also included with the package (rated, non-rated, line officers, etc.)

• Specification of ARIMA error model (differencing, degree of autoregression, seasonality, etc.) or allows for automatic ARIMA specification if desired

• Generation of model assessment statistics (RMSE, MAPE, residual analysis, graphical analysis)

• Graphic generation for presentation and evaluation purposes

The latest version of this package will be found at my github page. All military personnel data is unclassified and cleared for public release.

Progress Check

| Feature | Priority | Status | Value | Inputs | Outputs | Timeline | | ------------------- | -------- | ------ | ----------------------- | ----------------- | --------------------------------- | ------------------------------------ | | Variable selection | 1 | ns | model specification | selected vars | selected vars | sufficient time for current deadline | | ARIMA specification | 3 | ns | model specification | ARIMA attributes | selected model attributes | sufficient time for current deadline | | Personnel data | 2 | ns | model specification | personnel subset | selected subset | sufficient time for current deadline | | Assessment stats | 4 | ns | ID best fit model | stat selection | residual plots, accuracy measures | sufficient time for current deadline | | Graphics | 5 | ns | presentation/evaulation | graphic selection | forecast plots over actual data | sufficient time for current deadline |

jtelliott/attrition-forecast documentation built on May 4, 2019, 10:55 a.m.