Forecasting Unemployment Rates in the UK and EU Team B9 Rajesh Manivannan (61710602) Kartick B Muthiah (61710764) Debayan Das (61710492) Devesh Kumar (61710353) Chirag Bhardwaj (61710812) Sreeharsha Konga (61710646)
BUSINESS PROBLEM Governments need to allocate budgets for unemployment benefits as part of their social welfare schemes Unemployment: Are a huge cost to the government Unemployment definitions vary from country to country Each year the government allocate a certain percentage of their financial outlay Unemployment is generally considered as a lagging indicator of business cycles, however increase in unemployment has preceded the last 3 recessions Client Information: Our Clients are Ministries of Finance of European and UK governments who have to budget for unemployment benefits as part of their social welfare schemes.
DATA DESCRIPTION Seasonally unadjusted Unemployment data for each country is taken for Analysis Source: Federal reserve of Economic Data (https://fred.stlouisfed.org/) Key Characteristics: Trend, Level and noise observed for all the data. Seasonality observed for certain data. Countries to be analyzed: Austria, UK, Ireland, Germany, Poland, Luxembourg
DATA DESCRIPTION Each Unemployment Time Series is unique and hence various methods have to be applied Trend Series - Austria
DATA DESCRIPTION Each Unemployment Time Series is unique and hence various methods have to be applied Time Series - Germany
DATA DESCRIPTION Identifying Trend and Level Trend Series - Austria For Trend and Level: Regular time-series plot with a trend line For Seasonality: Yearly: X axis: Years; Y axis: Unemployment Rate Monthly: X Axis: Months; Y axis: Unemployment Rate; Plot each year as a line
DATA DESCRIPTION Forecasting process Identifying seasonality Seasonality Graph For Trend and Level: Regular time-series plot with a trend line For Seasonality: Yearly: X axis: Years; Y axis: Unemployment Rate Monthly: X Axis: Months; Y axis: Unemployment Rate; Plot each year as a line
FORECASTING PROCESS Data Partitioning was done using a validation period of 15 months Fiscal Year extends from Jan to Dec The client requires the forecast 3 months prior to the start of the Fiscal year Data set includes monthly Unemployment rate from Jan 1990 Sept 2014 Forecasting Horizon -> 15 months Seasonality -> 12 months Training period of 282 months and Validation period of 15 months was chosen
FORECASTING PROCESS Choosing the right method for Forecasting The following methods were used to find the Validation period MAPE Naive Smoothing o Exponential o Double Exponential o Moving Average (2) o Holt-Winters (Multiplicative, Additive and NoTrend) Multiple Linear Regression o Only seasonal variables o Auto-Regressive o Log(Unemployment Rate) o Sqrt(Unemployment Rate) o Inverse(Unemployment Rate) Validation period MAPE for all methods used for UK
FORECASTING PROCESS Forecasting Process for UK Multiple Linear Regression Output variable: Log(Unemployment Rate) Input variables: Time Time^2 Time^3 Time^4 11 Monthly dummy variables (excluding September) Lag -1 Lag -2
FORECASTING PROCESS Forecasting process UK Unemployment forecast values
FORECAST PROCESS Forecasts for Austria - 1
FORECAST Forecasts for Austria - 2 Output variable: Unemployment Rate Input variables: Time Time^2 11 Monthly dummy variables (excluding September) Lag -1
LIMITATIONS Challenges in predicting external economic indicators constraining the model for better forecast The MLR model uses Lag-1 as an input variable Lag-1 data will not be available ahead of time Lets us forecast only one month at a time Alternatives: - Use lag-13 if Naïve forecast has reasonable MAPE - Use another model to forecast lag-1 No external economic Indicators were used. The model is highly dependent on the frequency of collection of unemployment data.
CONCLUSION Holt winter methods generally give good results and can be used as reference for further MLR methods Recommendations *Numbers are an estimate For most countries we tried many methods of which MLR with lag 1 was highly accuarate with MAPE of under 2% However, MLR is difficult and more costlly to implement. So, one can use Holt Winters to forecast as well. We recommend sensitizing the forecasted values using the confidence interval. This should help the government adjust for buffer allowances A sample sensitivity analysis presented in the slide Learnings Holt Winter methods are quick and fairly accurate in their forecasts MLR using time index and seasonal dummy variables doesn t give better results than Holt Winters To improve MLR forecasts is using MLR + ARIMA method. However, this would required one input variable is lag 1 which limits the model s prediction capability
APPENDIX Forecast - Luxembourg
APPENDIX Forecast - Netherlands
APPENDIX Forecast - Poland