Macroeconomic News Announcement Effects on Stocks

1
Macroeconomic News Announcement Effects on Stocks

• Allison Keane

2
Motivation

• Determine if there exists a relationship between
  news announcements and stock returns
• News announcements occur before market opens
  need appropriate measure of return
• Need appropriate measure of standardization

3
Equations

• Returns
• R1000 log(P1000,t1) - log(Pclose,t )
• Announcements (data taken from Yahoo Finance)
• Skt (Akt Ekt ) / sk
• Have to standardize because of units
• Realized Variance
• RV Sr2 (calculated using five minute log
  returns)
• DRVt v RVt-1
• WRVt v((1/5)(RVt-1 RVt-2 RVt-3 RVt-4
  RVt-5))

4
Equations

• Standarizing R
• Standardize R since S is standardized
• Rt/DRVt
• Rt/WRVt
• Rt/MRVt
• Regressions
• Rt/WRVt ßkSk,t et

5
Different standardizations of R
6
Stocks

• Focused on four stock from SP100
• Begin with different industries
• Procter Gamble (PG)
• Kraft (KFT)
• American International Group (AIG)
• Ford (F)
• Later expand to see if similarities among
  industries
• Avon Product Inc.(AVP)
• Hartford Financial Group(HIG)
• Allstate Corp. (ALL)
• Colgate-Palmolive(CL)
• Data sets include one minute price data from 2002
  - 2007

7
Announcements

• Produce Price Index (PPI)
• Consumer Price Index (CPI)
• Durable Goods (D)
• Industrial Production (I)
• Retail Sales (R)
• Average Work Week (AWW)
• Unemployment Rate (UR)
• Hourly Earnings (HE)
• Nonfarm Payrolls (NP)
• Capacity Utilization (CU)
• Business Inventories (BI)
• Personal Income (PI)

• All announcements occur before market opens
• Any days announcements did not occur or data was
  not available are disregarded

8
Regression I Test Different Standardizations

• Attempt to determine which standardization value
  for R was best
• None, DRV, WRV, MRV
• Took the 1000 return from four primary stocks
  standardized four different ways
• Regressed each standardized R against each
  announcement individually
• R1000/ DRVt ßkSk,t et
• 192 different regressions

9
Regression I Results

• Found P-values very different for different
  combinations
• Could have small p-values for PG and KFT, but AIG
  and F would have high values
• Not very many significant coefficients for any
  standardization

Significant Betas
Average R2 values

• Examined highest R2 values but no consistent
  pattern
• Looked at averages and used the best
  standardization based on the average

10
Regression II Test Different Return Values

• Question Which return should be used as
  overnight return measure?
• Want to account more market adjustment
• Assume market will adjust quickly
• Test 935, 940, 945, 950 955, 1000, 1010,
  1020, 1030, 1040, 1050, 1100, 1110, 1120,
  1130, 1230, 300
• Use later times, 1230 and 300 to show the
  announcement has had an effect by then
• Standardize each return by WRV based on previous
  regression results
• R1000 log(P1000,t1) - log(Pclose,t )
• R1000/WRV ßkSk,t et
• Do this regression for each return measure for PG
  only against each announcement individually
• 214 regressions

11
Regression II Results

• P-values varied depending on the announcement
• - Some announcements had very high p-values for
  all returns, some had smaller values
• - General trend smaller p-values in morning
  relative to those in the afternoon
• There was not one consistent return with the
  lowest P-value
• Most lowest p-values occurred between 935 and
  1000 and only one past 1100
• Focused on returns between 935 and 1000 and
  used 1000 because had lowest average R2
• Difficulty the coefficients would change sign
• - When regressed against HE, the first two
  returns had pos coefficients and the rest were
  negative
• - Occasionally, just one coefficient would change
  sign

12
Regression III Multivariate Regressions

• Real Activity
• Rt ßk(NP) t ßk(R) t ßk(I) t ßk(CU) t
  ßk(PI) t et
• Prices
• Rt ßk(CPI) t ßk(PPI) t et
• Investment
• Rt ßk(BI) t ßk(D) t et
• Employment
• Rt ßk(NP) t ßk(HE) t ßk(AWW) t ßk(UR) t
  et

13
Regression III Results
Regress (F-stats)
Newey - West

• Was hoping to see PG, AVP, CL to have similar
  significant regressions and AIG, HIG, ALL have
  similarities
• F-test for all announcement on PG becomes
  insignificant if take out BI

14
Extensions

• Perform regressions on individual stocks for ALL,
  AVP, CL, HIG
• Add more stocks from similar industries and
  SP500 data
• In process of determining if response varies with
  sign
• ?k ß0 ß1k,t Sk,t if Slt0
• ß2 ß3k,t Sk,t if Sgt0