Refer to Problem 16-30.
a. Use the linear trend model (without transformation) for the first 15 months and provide a cash balance forecast for month 16. Then make the t2 transformation and develop a new linear trend forecasting model based on months 1 through 15. Forecast the cash balance for month 16. Now compare the accuracy of the forecasts with and without the transformation. Which of the two forecast models would you prefer? Explain your answer.
b. Provide a 95% prediction interval for the cash balance forecast for month 16 using the linear trend model both with and without the transformation. Which interval has the widest width? On this basis, which procedure would you choose?
In exercise
Logan Pickens is a plan/build construction company specializing in resort area construction projects. Plan/ build companies typically have a cash flow problem since they tend to be paid in lump sums when projects are completed or hit milestones. However, their expenses, such as payroll, must be paid regularly. Consequently, such companies need bank lines of credit to finance their initial costs, but in 2009, lines of credit were difficult to negotiate. The data file Logan-Pickens contains month-end cash balances for the past 16 months.
a. Plot the data as a time-series graph. Discuss what the graph implies concerning the relationship between cash balance and the time variable, month.
b. Fit a linear trend model to the data. Compute the coefficient of determination for this model and show the trend line on the time-series graph. Discuss the appropriateness of the linear trend model. What are the strengths and weaknesses of the model?
c. Referring to part b, compute the MAD and MSE for the 16 data points.
d. Use the t2 transformation approach and recompute the linear model using the transformed time variable. Plot the new trend line against the transformed data. Discuss whether this model appears to provide a better fit than did the model without the transformation. Compare the coefficients of determination for the two models. Which model seems to be superior, using the coefficient of determination as the criterion?
e. Refer to part d. Compute the MAD and MSE for the 16 data values. Discuss how these compare to those that were computed in part c, prior to transformation. Do the measures of fit (R2, MSE, or MAD) agree on the best model to use for forecasting purposes?