Forecasting for Economics and Business
Due 22 May 2025 at 9:00 AM
Assignment
For this assignment, you will be forecasting Australian interest rates — specifically, the RBA Cash Rate Target. The time series used in your analysis and forecasting can be obtained using the read_rba and read_abs functions from the readrba and readabs packages, respectively (you will need to install those packages first). You will use the former to obtain the interest rate data (series ID: “FIRMMCRTD”) and the latter to obtain inflation data (series ID: “A128478318V”).
Note: Interest rate data are available at a daily frequency, going back to the 1990s. Inflation data are monthly, starting from September 2018. For your analysis, in both instances, use monthly series from September 2018 onward. The inflation series currently extends to March 2025, which will suffice for nearly all of your analysis. The only instance where you’ll need April 2025 inflation is when forecasting the May 2025 interest rate using the ARDL model. In that case, assume that April 2025 inflation is equal to March 2025 inflation.
Part 1
The first part of the assignment involves:
1. Checking for stationarity in the two time series, observed at a monthly frequency. (Note: you will need to convert the daily interest rate data to monthly by selecting one observation on a later date in the month) Use all available data for this step. In one or two sentences, document your findings. Additionally, present both the interest rate and inflation series in the same figure.
2. Generating a monthly series of one-step-ahead forecasts of the RBA cash rate for an out-of-sample period beginning in January 2022. Use a rolling window forecasting scheme to a range of competing models. Specifically, generate forecasts from the following models:
(i) Random walk model
(ii) Autoregressive model (select lag order using SIC, allowing up to four lags)
(iii) Autoregressive distributed lag (ARDL) model, using only the lags (i.e., do not include contempora-neous values) of inflation as an exogenous variable. Use SIC to determine both autoregressive and distributed lag orders, allowing up to four lags in each case; allow for the possibility of different lag order, that is, in your ARDL(p,q), p doesn’t need to be equal to q.
For AR and ARDL models, select the optimal lags only once using the first rolling window, and retain them throughout the forecasting exercise. In one or two paragraphs, describe the model selection process and your rationale.
3. Assessing forecast unbiasedness and efficiency for each of the three models. Use heteroskedasticity- and autocorrelation (HAC) consistent standard errors (hint: use the coeftest and vcovHAC functions from the lmtest and sandwich packages). In one or two sentences, summarize your findings. Present the results of your unbiasedness and efficiency tests in one table.
4. Comparing forecast accuracy across the three models using quadratic loss and the Diebold-Mariano test statistic, again using HAC standard errors. To keep things simple, ignore any nesting between models; that is, do not apply any adjustment term to the loss differentials, even if you believe it may be warranted. Describe your findings in one or two paragraphs. Present the Diebold-Mariano test results in one table.
Part 2
The second part of the assignment involves forecasting the May 2025 interest rate. Your point forecast will almost certainly be off, as the RBA typically adjusts rates in standard basis point increments (e.g., the actual value is likely to be 4.10%, 3.85%, or 3.60%, but unlikely to fall in between). Once the RBA announces the May rate (in the afternoon of 20 May 2025), compare the realized value to your forecast. Briefly discuss the size and direction of your forecast error, and any reasons that may explain it.
Submission
Upload both your report (as a PDF) and the R file with your project code by the due date. The report should include one figure and two tables in addition to the text (which should not exceed 1500 words). The R file must be free of coding errors. Data should be sourced using the read_rba and read_abs functions to ensure full reproducibility of your results.