Title: Econ 240 C
1Econ 240 C
2Outline
- I. Transcendental ideas
- II. Lab Two
- III. Analysis and synthesis
- IV. Characterizing time series behavior
- V. Analysis in the Lab Process
3Ideas to remember
- Models are just a guide to thinking, not an end
in themselves - If the model is the box, think outside of the
box compare competing models - Sell complexity, believe simplicity
4Outline Lab Two
- Linear Trend model for sp500
- Exponential trend model lnsp500
- Other models
- Exponential trend plus random walk
- Naïve models
5Linear trend model
- Process combine what you learned in 240A, B, C
- Spreadsheet
- Plot or trace not linear
- Histogram not normal
- Correlogram not orthogonal
- Unit root test not stationary
- Sp500(t) a b t e(t)
- Goodness of fit
- Violation of regression assumptions
- One period ahead forecast from the trend model
6(No Transcript)
7(No Transcript)
8(No Transcript)
9(No Transcript)
10(No Transcript)
11(No Transcript)
12Linear trend model
- Process combine what you learned in 240A, B, C
- Spreadsheet
- Plot or trace not linear
- Histogram not normal
- Correlogram not orthogonal
- Unit root test not stationary
- Sp500(t) a b t e(t)
- Goodness of fit
- Violation of regression assumptions
- One period ahead forecast from the trend model
13Diagnostics R2 0.72, low for time
series D-W0.016, very low
14(No Transcript)
15Residuals are not orthogonal
16Residuals are not normal
17(No Transcript)
18Diagnostics R2 0.72, low for time
series D-W0.016, very low
19One period ahead forecast
- E2003.02 Sp500(2003.03) -772.7503
9.170486398 - E2003.02 Sp500(2003.03) -772.7503 3649.8534
- E2003.02 Sp500(2003.03) 2877.103
- Approximate 95 confidence interval /- 2ser
- Upper bound 2877.1 2651.4 4179.9
- Lower bound 2877.1 2651.4 1574.3
20(No Transcript)
21E2003.02 sp500(2003.03) Fitted(2003.02)
slope 2867.93 9.17 2877.1
E
22(No Transcript)
23(No Transcript)
24(No Transcript)
25Exponential (log-linear) trend model
- Process combine what you learned in 240A, B, C
- Spreadsheet
- Plot or trace more log-linear
- Histogram not normal
- Correlogram not orthogonal
- Unit root test not stationary
- Lnsp500(t) a b t e(t)
- Goodness of fit
- Violation of regression assumptions
- One period ahead forecast from the trend model
26(No Transcript)
27(No Transcript)
28(No Transcript)
29(No Transcript)
30Exponential (log-linear) trend model
- Process combine what you learned in 240A, B, C
- Spreadsheet
- Plot or trace more log-linear
- Histogram not normal
- Correlogram not orthogonal
- Unit root test not stationary
- Lnsp500(t) a b t e(t)
- Goodness of fit
- Violation of regression assumptions
- One period ahead forecast from the trend model
31R2 0.97, better fit D-W 0.042, very low
32(No Transcript)
33Residuals are not orthogonal
34Residuals are not normal
35Dependent Variable LNSP500 Method Least
Squares Sample(adjusted) 197001
200302 Included observations 398 after
adjusting endpoints Variable Coefficient Std
. Error t-Statistic Prob. C 4.049837 0.02
2383 180.9370 0.0000 TIME 0.010867 9.76E-05 111.3
580 0.0000 R-squared 0.969054 Mean
dependent var 6.207030 Adjusted
R-squared 0.968976 S.D. dependent
var 1.269965 S.E. of regression 0.22368
Akaike info criterion -0.152131 Sum squared
resid 19.8141 Schwarz criterion -0.132099 Log
likelihood 32.27406 F-statistic 12400.61 D
urbin-Watson stat 0.041769 Prob(F-statistic) 0.
000000
36Fitted model
- Lnsp500(t) c bt
- 2003.03 is t398
- Forecast lnsp500(t398) 4.049837
3980.010867 - Forecastlnsp500(2003.03) 8.374903
- Upper bound of 95 confidence interval
8.374903 2 SER 8.374903 20.22368 - Upper bound 8.822263
37Fitted Model
- Lower bound forecast -2ser 8.374903 -0.44736
7.927543
38(No Transcript)
39Fitted model
- Forecast of sp500(2003.03) exp(8.374903)
4336.848 - Upper bound exp(8.822263) 6783.599
- Lower bound exp(7.927543) 2772.606
40(No Transcript)
41Alternative models
- Sp500(t) expa bt rw(t)
- Lnsp500(t) a bt rw(t)
- ?lnsp500(t) b ? rw(t)
- ?lnsp500(t) b wn(t)
420.00862512 0.1035 annual rate of growth
43Residuals are orthogonal
44(No Transcript)
45Mean of dlnsp500, H0 ? 0, 0.008625 0/std
dev/n1/2 Students t-statistic 3.76
46Alternative models
- Naïve forecast best forecast of next period is
this periods value - True for a random walk
- Lnsp500(387) 7.86996118907
- Exp lnsp500(387) 2617.464
47(No Transcript)
48Conclusion
- Best model lnsp500(t) a bt rw(t)
- Best long run forecast is trend a bt
- Best short run forecast is random walk rw(t)
49 III Box-Jenkins Magic
- ARMA models of time series all built from one
source, white noise
50Analysis and Synthesis
- White noise, WN(t)
- Is a sequence of draws from a normal
distribution, N(0, s2 ), indexed by time
51Analysis
- Random walk, RW(t)
- Analysis formulation
- RW(t) RW(t-1) WN(t)
- RW(t) - RW(t-1) WN(t)
- RW(t) ZRW(t) WN(t)
- 1 ZRW(t) WN(t)
- DRW(t) WN(t) shows how you turn a random walk
into white noise
52Synthesis
- Random Walk, Synthesis formulation
- RW(t) 1/1 ZWN(t)
- RW(t) 1 Z Z2 .WN(t)
- RW(t) WN(t) ZWN(t) .
- RW(t) WN(t) WN(t-1) WN(t-2) shows how
you build a random walk from white noise
53Analysis
- Autoregressive process of the first order,
analysis formulation - ARONE(t) bARONE(t-1) WN(t)
- ARONE(t) - bARONE(t-1) WN(t)
- ARONE(t) - bZARONE(t) WN(t)
- 1 bZARONE(t) WN(t) is a quasi-difference
and shows how you turn an autoregressive process
of the first order into white noise
54Synthesis
- Autoregressive process of the first order,
synthetic formulation - ARONE(t) 1/1 bZWN(t)
- ARONE(t) 1 bZ b2Z2 .WN(t)
- ARONE(t) WN(t)bZWN(t)b2Z2 WN(t) ..
- ARONE(t) WN(t) bWN(t-1) b2WN(t-2) .
Shows how you turn white noise into an
autoregressive process of the first order
55Part IV Characterizing Time Series Behavior
- Mean function, m(t) E time_series(t)
- White noise m(t) E WN(t) 0, all t
- Random walk m(t) EWN(t)WN(t-1) .. equals
0, all t - First order autoregressive process,
m(t) EWN(t) bWN(t-1) b2WN(t-2)
equals 0, all t - Note that for all three types of time series we
calculate the mean function from the synthetic
expression for the time series.
56Characterization the AutocovarianceFunction
- EWN(t)WN(t-u) 0 for ugt0 , uses the
orthogonality (independence) property of white
noise - ERW(t)RW(t-u) EWN(t)WN(t-1) WN(t-2)
WN(t-u)WN(t-u-1) s2 s2 s2 ....
, uses the orthogonality property for
white noise plus the theoretically infinite
history of a random walk
57The Autocovariance Function
- EARONE(t)ARONE(t-u) bEARONE(t-1)
ARONE(t-u) EWN(t)ARONE(t-u) - gAR,AR(u) b gAR,AR(u-1) 0 ugt0, uses both
the analytic and the synthetic formulations for
ARONE(t). The analytic formulation is used to
multiply by ARONE(t-u) and take expectations. The
synthetic formulation is used to lag and show
ARONE(t-1) depends only on WN(t-1) and earlier
shocks.
58The Autocorrelation Function
- rx,x(u) gAR,AR(u)/ gAR,AR(0)
- White Noise rWN,WN(u) 0u
- Random Walk rRW,RW(u) 1, all u
- Autoregressive of the first order rx,x(u) bu
59Visual Preview of the Autocorrelation Function
60Visual Preview of the Autocorrelation Function
61Visual Preview of the Autocorrelation Function
62Drop Lag Zero The Mirror Image of the Mark of
Zorro
Random Walk
1
First Order Autoregressive
White Noise
0
Lag
63Part V .Analysis in the Lab Process
- Identification
- Estimation
- Verification
- Forecasting
64Analysis in the Lab Process
- Identification
- Is the time series stationary?
- Trace
- Histogram
- Autocorrelation Function
- If it is, proceed
- If it is not, difference (prewhitening)
65(No Transcript)
66(No Transcript)
67(No Transcript)
68(No Transcript)
69Process Analysis in the LAB
- Identification
- conclude evolutionary
- Fix-up pre-whiten by first differencing
70(No Transcript)
71(No Transcript)
72(No Transcript)
73(No Transcript)
74(No Transcript)
75Identification
- Conclude it is stationary
- Conjecture ARONE Model
76Process Analysis in the LAB
- Estimation
- EVIEWS model
- time series(t) constant residual(t)
- residual(t) bresidual(t-1) WN(t)?
- Combine the two
- time series(t) - c btime series(t-1) - c
WN(t)? - EVIEWS Specification
- dinvsratio c ar(1)
77(No Transcript)
78(No Transcript)
79Estimation
- Goodness of Fit
- Structure in the residuals? Are they orthogonal?
- Are the residuals normally distributed?
80(No Transcript)
81(No Transcript)
82(No Transcript)
83Process Analysis in the LAB
- Identification
- Estimation
- Verification
- Is there any structure left in the residuals? If
not, we are back to our building block,
orthogonal residuals, and we accept the model.
84Process Analysis in the LAB
- Identification
- Estimation
- Verification
- Is there any structure left in the residuals? If
not, we are back to our building block,
orthogonal residuals, and we accept the model. - Forecasting
- one period ahead forecasts
85Process Analysis in the LAB
- Forecasting
- The estimated model
- dinvsratio(t) (-0.001430)
-0.244617dinvsratio(t-1) 0.001430 WN(t)
where WN(t) is approximately an independent error
series and is normally distributed - The forecast is based on the estimated model
- dinvsratio(2007.02) 0.001430
-0.244617dinvsratio(2007.01) 0.001430
WN(2007.02)
86Process Analysis in the LAB
- Estimation
- EVIEWS model
- time series(t) constant residual(t)
- residual(t) bresidual(t-1) WN(t)?
- Combine the two
- time series(t) - c btime series(t-1) - c
WN(t)? - EVIEWS Specification
- dinvsratio c ar(1)
87The Forecast
- Take expectations of the model, as of 2007.1
- E2007.01 dinvsratio(2007.02) 0.001430
-0.244617E2007.01 dinvsratio(2007.01)
0.001430 E2007.01 WN(2007.02) - E2007.01 dinvsratio(2007.02) is the forecast
conditional on what we know as of 2007.01 - dinvsratio(2007.01) 0.02, the value of the
series in 2007.01 - E2007.01 WN(2007.02) 0, the best guess for the
shock
88The Forecast
- Calculate the forecast by hand
- for a one period ahead forecast, the standard
error of the regression can be used for the
standard error of the forecast - calculate the upper band forecast 2SER
- calculate the lower band forecast - 2SER
89The Forecast
90(No Transcript)
91(No Transcript)
92(No Transcript)