Each student will complete an empirical research paper on an economic variable of your choice. The paper will help reinforce the econometric methods discussed in class. The class topics build upon themselves and the paper will follow a similar structure. This is an empirical exercise to demonstrate your ability to build and forecast using various time-series models.
Paper Structure:
1. Introduction
a. Statement of the topic and question to be analyzed
b. Rational for choosing the topic
c. Outline remainder of paper (What will be covered in each section)
2. Body
a. Literature Review
b. Application of economic theory
c. Analysis of data
d. Results
3. Conclusion
4. References (Use Chicago/Turabian format)
Forecasting Paper Hints
Data Analysis: Descriptive statistics, graphs, histogram, unit-root tests, transformations (logs, differences etc), any seasonal complications?
? Discuss the main series you are forecasting. What are its statistical properties? Any difficulties with the series? Personal thoughts on what kind of trend you think it should have, if any?
? Show tables, graphs etc to support your preliminary data analysis
Univariate Models: ARMA model, TS, DS
? Why is it useful to start with simple autogressive models?
? Fit the best ARMA model. Forecast in-sample and out sample to see who well it performs.
? Do you want to estimate a TS or DS or both? What did your unit root tests tell you? Do you believe them? Forecast in and out of sample. Compare the TS to the DS and to the ARMA. Which is the best? Any thoughts why?
? Remember to use a hold-out sample. If you have data through 2012 and its monthly maybe try leaving a year or even just 6 months so you can see how you forecast in sample AND out of sample.
Multivariate Models: ARDL models, Vector auto-regressions, Error-correction models
? Are there any other predictor variables you want to add to your best TS or DS model to make it an ARDL model? How many lags to include?
? You will need to produce a forecast for each of those other independent variables. Should you just fit a simple univariate model for them? Want to make it more complicated? Solve using the model module.
? How many variables would you want to include in a vector autoregression (They use up degrees of freedom)? Why estimate a VAR over a simple multivariate model? Does the Granger Causality test support your thoughts? Impulse response functions? How do the in-sample and out-of-sample forecasts compare to other models?
? Co-integration tests for long-run relationships between two stochastic trending variables. Do you think 2 or more of your variables share this type of relationship? Can you estimate a Vector-error correction model instead of a simple VAR? Does it help you forecast better? Compare once again.
Use the order calculator below and get started! Contact our live support team for any assistance or inquiry.
[order_calculator]