Data Analysis II-Derrick Hang 2.24.09

advertisement
Derrick Hang
February 24, 2010
Economics 201FS
Apple Inc. (AAPL):
April 16, 1997 – January 7, 2009
2,920 Days
IBM (IBM):
April 9, 1997 – January 7, 2009
2,925 Days
Proctor Gamble Co (PG):
April 9, 1997 – January 7, 2009
2,924 Days *
*Absence of a day from PG was found to be on March 7, 2000; this is when PG
released a warning that earnings for the rest of the fiscal year will fall short of
expectations, contributing 142 to the 374 point drop in the Dow that day
• 8 minute intervals are used so the intervals fit exactly
with the 385 minutes per day window; in other words,
there is no incomplete interval left over at the end of
each day
i.e. to obtain the first 8-minute return you take the price
level at the 1st min. and the 9th min., then the 9th and the 15th,
etc.; following this sequence, we end up using the price
level at the 385th and 377th min.
•
• 8 minute is consistent with the volatility signature
plots presented last time, and leaves no incomplete
interval at the end
0.1%
1%
5%
IBM
Significance Level = 75 / 2925 (2.56%)
Significance Level = 219 / 2925 (7.49%)
Significance Level = 539 / 2925 (18.43%)
0.1%
1%
5%
AAPL
Significance Level = 113 / 2920 (3.87%)
Significance Level = 305 / 2920 (10.45%)
Significance Level = 625 / 2920 (21.40%)




PG
Significance Level = 82 / 2924 (2.80%)
Significance Level = 223 / 2924 (7.63%)
Significance Level = 548 / 2924 (18.74%)




0.1%
1%
5%




0.1%
1%
5%
IBM
Significance Level = 80 / 2925 (2.74%)
Significance Level = 236 / 2925 (8.07%)
Significance Level = 559 / 2925 (19.11%)
0.1%
1%
5%
AAPL
Significance Level = 136 / 2920 (4.66%)
Significance Level = 326 / 2920 (11.16%)
Significance Level = 653 / 2920 (22.36%)




PG
Significance Level = 94 / 2924 (3.21%)
Significance Level = 238 / 2924 (8.14%)
Significance Level = 563 / 2924 (19.25%)




0.1%
1%
5%







0.1% Significance Level = 65 / 2920 (2.23%)
1%
Significance Level = 143 / 2920 (4.90%)
5%
Significance Level = 328 / 2920 (11.23%)



0.1% Significance Level = 57 / 2925 (1.95%)
1%
Significance Level = 126 / 2925 (4.31%)
5%
Significance Level = 278 / 2925 (9.50%)



0.1% Significance Level = 61 / 2924 (2.09%)
1%
Significance Level = 143 / 2924 (4.89%)
5%
Significance Level = 285 / 2924 (9.75%)
• Bayesian Forecasting with Dynamic Models
using high-frequency data
• Regression where every variable varies with time
• Better coefficients from use of high frequency
data? What time window has better
predictability?
• What should be the dependent: returns, prices,
volatility?
Download