Processing math: 100%

  1. Bullet points. These are appropriate in presentations, but are best avoided in reports. Why?

  2. Code. Generally, code should be copied into a report only when the specifics of coding are a topic of discussion. My notes are not the best role model for a report: I include plenty of code, in order to help discussion of the process of generating the analysis.

  3. Fitted values. The so-called fitted values for ARMA models are the one-step forecasts.

    • These can look like a good match for the data even when the model has no predictive power beyond simply estimating the next data point using the current data point.

    • Usually, it is not interesting to note that a simple forecast (using the previous data point or the previous year at the same time) looks visually like a good match for the data.

    • It may be more interesting to note whether the fitted model does substantially better than a trivial forecast.

  4. Taking logs. Remember to consider the possibility of transforming the data, usually a square root or log transform for non-negative data.

  5. Complex conjugate ARMA roots. Several people noticed that roots had the form a±bi, which is not statistically interesting since it is just the form of complex roots of a quadratic.