Presentation is loading. Please wait.

Presentation is loading. Please wait.

1 Power Nine Econ 240C. 2 Outline Lab Three Exercises Lab Three Exercises –Fit a linear trend to retail sales –Add a quadratic term –Use both models to.

Similar presentations


Presentation on theme: "1 Power Nine Econ 240C. 2 Outline Lab Three Exercises Lab Three Exercises –Fit a linear trend to retail sales –Add a quadratic term –Use both models to."— Presentation transcript:

1 1 Power Nine Econ 240C

2 2 Outline Lab Three Exercises Lab Three Exercises –Fit a linear trend to retail sales –Add a quadratic term –Use both models to forecast 1 period ahead Lab Five Preview Lab Five Preview –Airline passengers

3 3 Lab Three Exercises Process Identification Identification –Spreadsheet –Trace –Histogram –Correlogram –Unit root test Estimation Estimation Validation Validation

4 4 1955.01-1993.12

5 5

6 6

7 7

8 8

9 9

10 10 Lab Three Exercises Process Identification Identification –Spreadsheet: check variable values –Trace: trended series –Histogram: similar to “random walk” –Correlogram: similar to a “random walk” –Unit root test: evolutionary Estimation Estimation Validation Validation

11 11 Note: low D-W

12 12 Process Validating the model Validating the model –Actual, fitted, residual –Correlogram of the residuals –Histogram of the residuals

13 13 Note: autocorrelated residuals

14 14 Note autocorrelated residuals

15 15 Surprise: residuals are normal, but not orthogonal

16 16 Add the quadratic term

17 17 SER is lower: 3737 Vs. 4860, D-W still low

18 18 Autocorrelated residual

19 19 Autocorrelated residual

20 20 Residual is no longer normal

21 21 One Period Ahead Forecasts: Linear Retail(1994.01) = retail-fitted(1993.12) + slope Retail(1994.01) = retail-fitted(1993.12) + slope Forecast = 140339 + 215.0 = 140553, with ser = 4860 Forecast = 140339 + 215.0 = 140553, with ser = 4860

22 22

23 23

24 24

25 25 One Period Ahead Forecast: Quadratic Retail(t) = a + b*trnd + c*trnd 2 Retail(t) = a + b*trnd + c*trnd 2 d retail(t)/d trend = b + 2*c*trnd = 126.2 + 2*0.190*468 = 304.0 d retail(t)/d trend = b + 2*c*trnd = 126.2 + 2*0.190*468 = 304.0 Retail(1994.01) = retail-fitted(1993.12) + 304 = 147,239 + 304 = 147543 with ser = 3737 Retail(1994.01) = retail-fitted(1993.12) + 304 = 147,239 + 304 = 147543 with ser = 3737

26 26 SER is lower: 3737 Vs. 4860, D-W still low b = 128.2, c = 0.19

27 27 Trnd(1994.01) = 468

28 28 Retail-fitted(1993.12) = 147,239

29 29

30 30

31 31 Now we know another way to forecast First difference retail First difference retail

32 32

33 33 Looks stationary

34 34 kurtotic

35 35 Possibly an ARTWO

36 36 No unit root

37 37 Note: the constant 221 is close to the Slope, 215, for the Linear trend model

38 38

39 39 Q-stats ok until Lag 16

40 40

41 41 One period ahead forecast Dretailf(1994.01) = -240.66 with ser of 1230 Dretailf(1994.01) = -240.66 with ser of 1230 Retailf(1994.01) = retail(1993.12) + dretailf(1994.01) Retailf(1994.01) = retail(1993.12) + dretailf(1994.01) Retailf(1994.01) = 151,631 – 240.7=151,390.3 Retailf(1994.01) = 151,631 – 240.7=151,390.3 Linear trend forecast: 140,553 with ser = 4860 Linear trend forecast: 140,553 with ser = 4860 Quadratic forecast: 147,543 with ser =3737 Quadratic forecast: 147,543 with ser =3737 Actual observed: retail(1994.01) = 149,918 Actual observed: retail(1994.01) = 149,918 So the ARTWO model is closest So the ARTWO model is closest

42 42 dretail ARTWO Model

43 43 dretail ARTWO model forecast

44 44 One period ahead forecast cont. Add an ar(1) term to the quadratic model: forecast(1994.01) = 151,763 with ser = 1252 Add an ar(1) term to the quadratic model: forecast(1994.01) = 151,763 with ser = 1252 So dretail is still closest So dretail is still closest Add ar(1) ar(2) ar(3) terms to the quadratic: forecast(1994.01) = 151,457 with ser = 1223; Add ar(1) ar(2) ar(3) terms to the quadratic: forecast(1994.01) = 151,457 with ser = 1223; Still closest with ARTWO model for dretail, but not by much Still closest with ARTWO model for dretail, but not by much

45 45 ARTHREE Model for Quadratic Trend model of Retail

46 46 ARTHREE Model for Retail

47 47 Residuals from Quad. trend model plus ARTHREE

48 48

49 49 Preview of Lab Five A Box-Jenkins famous time series: airline passengers A Box-Jenkins famous time series: airline passengers –Trend in mean –Trend in variance –seasonality Prewhitening Prewhitening –Log transform –First difference –Seasonal difference

50 50

51 51

52 52

53 53 Note trend from Spike in pacf at Lag one; seasonal Pattern in ACF

54 54

55 55

56 56 Log transform is fix for trend in Var

57 57 First difference for trend in mean Looks more stationary but is it?

58 58

59 59 Note seasonal peaks at, 12 24, etc.

60 60 No unit root, but Correlogram shows Seasonal Dependence on time

61 61

62 62

63 63

64 64 Note: sddlnbjpass is normal

65 65 Closer to white Noise; proposed Model ma(1), ma(12)

66 66

67 67

68 68 Satisfactory Model from Q-stats

69 69 And the residuals from the model are normal

70 70 How to use the model to forecast Forecast sddlnbjpass Forecast sddlnbjpass recolor recolor

71 71

72 72


Download ppt "1 Power Nine Econ 240C. 2 Outline Lab Three Exercises Lab Three Exercises –Fit a linear trend to retail sales –Add a quadratic term –Use both models to."

Similar presentations


Ads by Google