More explorations with the old QMM

Day 3 morning

1. Recap

   • Limitations of interactive clicking
   • Developing a workflow to answer a question

2. Exploration

   • Stochastic simulation
   • Ex post forecasting
   • Ex ante forecasting

Day 3 afternoon

2. Partially repeat on a bigger scale – New QMM

3. Special topics, e.g. output gap

What “solve the model” means

Deterministic Stochastic solution

\[\begin{eqnarray*} Y_t&=& C_t+I_t \\ C_t &=&\widehat{\beta}_0+\widehat{\beta}_1 Y_t \color{red}{+\widehat{u}_{C,t}} \\ I_t &=&\widehat{\gamma}_0+\widehat{\gamma}_1 I_{t-1} {\color{red}{+\widehat{u}_{I,t}}} \end{eqnarray*} \] Solve for \(Y_t\) and \(C_t\) at every point in time for period under consideration.

Picture from old QMM final report

1. Ex post forecasting is like pseudo-out-of-sample forecasting.

   • Split available data into training and test data.
   • Use test data for out-of-sample forecasting.

Picture from old QMM final report

2. Ex ante forecasting is actual out-of-sample forecasting.

   • No test data simply because they don’t exist yet.
   • Must fill up the required data.

3. The old QMM final report does not have results on either.

Conducting ex post forecasting

1. 2003q1-2003q4 data are available.

2. Find the baseline by solving the model.

3. Compare actual with baseline.

   • Probably not a good idea to report RMSE and RMSPE given the rather short data availability.
   • But, in principle, you can calculate these.
   • Better to draw graphs.

Conducting ex ante forecasting

1. Conceptual issues, practical issues
2. Consider a real sector version of old QMM for practice. Exclude the labor market.
3. Use the equations and identities from Section 3.1 of the QMM Final Report to create this version of QMM. Create a new model object.
4. Try solving the model for 2003q1 to 2007q4.
5. What happens?

Conducting ex ante forecasting

6. You have assumptions about what the future looks like. This becomes your baseline.
7. You have plans for the future. This becomes your scenario.
8. But how do these differ from Day 2?
9. Do you understand why these were left almost undocumented in the old QMM final report?
10. Can you imagine what these would look like in the new QMM?

Other questions raised

1. Sector-specific shocks in old QMM
2. Permanent and temporary shocks

Workflows

1. Data input, program, clean copy
2. Routine maintenance
3. Historical simulation, ex post forecasting, ex ante forecasting
4. Communication of results
5. Data updates, re-estimation and revision, updating past results
6. Multiple team involvement