Question 3 (a) Explain the differences between training dataset and testing dataset, and why they are important in the cross-validation process (10 marks). (b) A researcher with a market research firm collected data for monthly sales growth of national cars from January 2015 to December 2020. He used part of the data covering January 2015 to August 2020 to estimate an autoregressive model shown below: y=0.21+0.15 yt-1-0.30 yt-2 The actual data for August to December 2020 (5 observations) were not used to estimate the model, but were kept aside for out-of-sample forecasting. The model estimates were used to forecast sales growth over the August-December 2020 period. The forecasts and actual sales figures are compared in the table below: Forecast horizon 2016 2017 2018 2019 2020 Actual GDP growth Forecast of GDP growth -0.55 0.23 0.27 0.17 0.13 0.11 -0.12 0.05 0.02 -0.03 Calculate (i) Mean absolute error (5 marks) (ii) Mean squared error (5 marks) (iii) Mean absolute percentage error (5 marks)