Comment on: Automated Short-Run Economic Forecast (ASEF) By Nicolas Stoffels. Bank of Canada Workshop October 25-26, 2007

Transcription

1 Background material Comment on: Automated Short-Run Economic Forecast (ASEF) By Nicolas Stoffels Bank of Canada Workshop October 25-26, 2007 André Binette (Bank of Canada) 1

2 Summary of ASEF 1. Automated process 2. Assess economic activity from numerous and timely indicators 3. Large set of forecasting equation Combination of forecasts and assessment of risk 4. Main advantages Fast, timely, flexible, ensure consistency and not costly to operate 5. Main disadvantages Difficult to track down the macro story 2

3 Monitoring Canadian Economic Development 1. Forecast (2-quarters-ahead) and follow the main economic variables Variables: GDP and its components, inflation and labour market variables Each cycle begins with the quarterly NAC publication Limited use of models, especially at the disaggregated level Mostly a judgment-based forecast driven by the data and an overall assessment of current economic conditions 2. As monthly data for the current quarter are published, we reassess our forecast (nowcasting) On a monthly basis or following Monetary Policy needs Mostly based on judgment + some mapping equations 3

4 Monitoring Canadian Economic Development 1. This judgment based approach has been used for several years 2. Still staff face some challenges Time consuming approach (80%-90%) Level of details and frequency of monitoring update Leaves little time for interesting and needed analysis on current economic issues Labour intensive (6 economists + 1 principal researcher) Hard to maintain the consistency: interpretation of the data and evolution of changes Very difficult for junior staff: no systematic approach 4

5 Time for a change 1. Currently we are working on developing an automated process which hopefully will resolve some of the challenges faced by our group 2. Key elements of our project: A. Core quarterly forecasting model for the main variables of interest B. Nowcasting equation for the main variables of interest Indicators Monthly dynamics C. Forecast comparison and evaluation Best combination at any point in time given the flow of data D. Global evaluation New vs. old approach E. Real time data F. Other models 5

6 A. Core quarterly forecasting model ECM: Main variables: real interest rate, real income and different measures of wealth Model structure chosen based on in-sample fit (adjusted R 2 around 64%) Sample: 1982:1 to 2007:2 Out-of-sample forecast accuracy Model re-estimated every quarter (extended window) RMSE and hit ratio calculated over 2001Q1-2007Q2 Evaluation RMSE = 1.7% (q/q at A.R.) Hit ratio = 73% 6

7 B. Nowcasting equation 1 Indicators Quarterly forecast based on 6 indicators (retail trade, car sales, travel, recreation activity, accommodation and food, and weather) Rolling window of 10 years (adjusted R 2 around 70%) Monthly forecast: 3 months average Could be improved Evaluation 0 month 1 month 2 month 3 month RMSE Hit ratio 42% 46% 62% 77% 7

8 B. Nowcasting equation 2 Monthly dynamic Quarterly forecast based on snapshot (assuming t-1 level for the remaining month/months of the quarter) Again forecast accuracy improved with more information 0 month 1 month 2 month 3 month RMSE Hit ratio 35% 58% 77% --- 8

9 C. Forecast comparison and evaluation Which model or combination gives the best RMSE (and hit ratio) at any point in the quarter? 1. Core model 2. Indicators model 3. Core model + indicators model 4. Core model + snapshot + indicators model 9

10 Q1 01Q3 02Q1 02Q3 03Q1 03Q3 04Q1 04Q3 05Q1 07Q1 0 01Q1 01Q3 06Q3 07Q1 Cons. 05Q3 06Q1 06Q3 Core model 02Q1 02Q3 03Q1 03Q3 04Q1 04Q3 Cons. 05Q1 05Q3 06Q1 Indicators model Q1 01Q3 02Q1 02Q3 03Q1 03Q3 Cons. 04Q1 04Q3 05Q1 05Q3 06Q1 Core + indicators 06Q3 07Q1 0 01Q1 01Q3 02Q1 02Q3 03Q1 Cons. 03Q3 04Q1 04Q3 05Q1 05Q3 06Q1 06Q3 07Q1 Core + indicators + snapshot 10

11 C. Forecast evaluation (RMSE) 2,2 2 1,8 1,6 1,4 1,2 0,8 1 0,6 0 month 1 month 2 month 3 month Core model Indicators model Core + indicators Core + indicators + snapshot 11

12 C. Forecast evaluation (hit ratio) 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0 month 1 month 2 month 3 month Core model Indicators model Core + indicators Core + indicators + snapshot 12

13 C. Forecast combination: model weights month 1 month 2 month 3 month Core model Snapshot Indicators model 13

14 Q1 02Q2 02Q3 02Q4 03Q1 03Q2 03Q3 03Q4 04Q1 04Q2 04Q3 04Q4 05Q1 05Q2 05Q3 05Q4 06Q1 06Q2 06Q3 06Q4 07Q1 07Q2 Cons. Core model Core + indicators + snapshot 90% interval 14

15 Conclusion Advantages of the new approach: Faster and timely Story easy to tell (fundamentals + monthly indicators) Systematic approach Less judgment needed Make our judgment explicit = we can evaluate it Leaves more time for analyzing others economic issues Improves the forecast accuracy throughout the quarter Reduces the RMSE by roughly half Increases the hit ratio by 15 ppt 15

16 Future Work 1. Are we gaining in terms of forecast accuracy? New vs. old approach (judgment based) 2. Should also be done with Real time and survey data 3. All this will be done for GDP and his main components Aggregate approach vs. component approach 4. Improving monthly forecasts 16