The Market Impact of Trends and Sequences in Performance:

1
The Market Impact of Trends and Sequences in
Performance New Evidence by Greg Durham Mike
Hertzel Spencer Martin College of Business,
W.P. Carey School W.P. Carey School Montana
State of Business, Arizona of Business,
Arizona University State University State
University
2
The Market Impact of Trends and Sequences in
Performance New Evidence FROM THE COLLEGE
FOOTBALL WAGERING MARKET by Greg Durham Mike
Hertzel Spencer Martin College of Business,
W.P. Carey School W.P. Carey School Montana
State of Business, Arizona of Business,
Arizona University State University State
University
3
Empirical Pricing Anomalies

• Momentum over Shorter Horizons
• Indexes Butler, Poterba, Sum- mers (1991)
• Stocks Jegadeesh Titman (1993), and others
• Reversals over Longer Horizons
• DeBondt Thaler (1985),
• and others

4
Behavioral Models

• Daniel, Hirshleifer, Subrahmanyam (1998)
• Self-Attribution Bias and Overconfidence
• Hong Stein (1999)
• Bounded Rationality
• of particular interest to this study is
• Barberis, Shleifer Vishny (1998)
• Conservatism Bias
• Reliance on the Represen- tativeness
  Heuristic

5
BSVs Regime-Shifting Model

• Conservatism Bias (Edwards, 1968)
• Underestimation of the value of new information
• Over-reliance on older information
• Representativeness Bias (Tversky Kahneman,
  1974)
• Over-reliance on similarities to the parent pop-
• ulation and on the salient features
• of an event
• Insufficient regard to other
• important factors

6
BSVs Regime-Shifting Model

• In actuality, a firms earnings performance fol-
  lows a random walk ... yet, investors believe
  that performance switches between
• Continuation (or Trending) Regime
• performance tends to be followed by like
  performance
• Reversal Regime
• performance tends to reverse
• i.e., returns are mean-reverting

7
Testable Implications

• The nature of the relation between prior
  performance and current prices turns out to be
  a key testable implication of the model
  developed by BSV
• Performance follows a random walk
• In formulating beliefs, investors examine past
  perfor- mance

8
Sports Betting Markets

• Bettors have real wealth at stake
• Numerous parallels to securities markets
• ? Informed bettors ? Experts
• Sentiment bettors ? Market makers
• Point spreads are used to balance books
• A sports bet has an obvious settling up
  point, at which terminal payoffs are
  unambig- uously realized

9
College Football Wagering Dataset
8 seasons of games from Division I-A, 1991-98 For
each game ? Opening Spread ? Change in
Spread ? Closing Spread ? Actual
Outcome Purchased from Computer Sports
World Spreads posted by Las Vegas Stardust
Casinos Sports Book
10
Point-Spread Market Mechanics

• Games almost always occur on Saturdays
• Betting begins Sunday night prior
• Odds and cash flows are fixed, so market makers
  quote point spreads
• Investors pay 11 to win 21 or 0
• For each pair of 11 bets on each team, 21 is
  paid ? transaxn costs 4.54
• Spreads fluctuate during week, but the
  expected change, in an efficient
  market, is zero

11
Mechanics Demonstrated

• Spreads fluctuate during week in response to an
  imbalance of orders (wagers) on one team

MSU v. UofM
MSU ?5
kickoff
12
Mechanics Demonstrated

• Spreads fluctuate during week in response to an
  imbalance of orders (wagers) on one team

MSU v. UofM
MSU ?5
MSU ?5
kickoff
Sun.
13
Mechanics Demonstrated

• Spreads fluctuate during week in response to an
  imbalance of orders (wagers) on one team

MSU v. UofM
MSU wins by gt5!!
MSU ?5
MSU ?5
kickoff
Sun.
14
Mechanics Demonstrated

• Spreads fluctuate during week in response to an
  imbalance of orders (wagers) on one team

MSU v. UofM
MSU ?5
MSU ?5
MSU wins by lt5 or loses outright!!
kickoff
Sun.
15
Mechanics Demonstrated

• Spreads fluctuate during week in response to an
  imbalance of orders (wagers) on one team

MSU v. UofM
MSU ?5
MSU ?5
MSU wins by 5!!
kickoff
Sun.
16
Mechanics Demonstrated

• Spreads fluctuate during week in response to an
  imbalance of orders (wagers) on one team

MSU v. UofM
MSU ?5
Good NEWS for MSU!!
kickoff
Sun.
17
Mechanics Demonstrated

• Spreads fluctuate during week in response to an
  imbalance of orders (wagers) on one team

MSU v. UofM
MSU ?6.5
MSU ?5
Good NEWS for MSU!!
kickoff
Sun.
18
Now, the Tests

• Testable implications of the BSV model
• Performance follows a random walk
• In formulating beliefs, investors examine past
  perfor- mance

19
Does Performance Follow Random Walk?

• Sorting Observations by Streak Length Suggests
  Yes (Table I)
• Observations per bin fall by 50 with each
  successive increment in streak length
• Team-by-Team Runs Tests Suggest Yes (Table
  II)
• For 105 of 113 teams, num- ber of runs is
  normal

20
Random Walk?
21
Does Performance Follow Random Walk?

• Sorting Observations by Streak Length Suggests
  Yes (Table I)
• Observations per bin fall by 50 with each
  successive increment in streak length
• Team-by-Team Runs Tests Suggest Yes (Table
  II)
• For 105 of 113 teams, num- ber of runs is
  normal

22
RandomWalk?
23
Do Investors Use Recent Freqs of Reversals?

• Identified teams with same 16 patterns as used by
  Bloomfield and Hales (JFE, 2002) (Table III)
• ?Spread is insignificant for all groups
• Mean changes are not different across low-,
  medium-, high-reversal groups
• Findings are inconsistent with the
  experimental subject results
• Findings are inconsistent with predictions of
  the BSV Model

24
Do Investors Use Recent Freqs of Reversals?
A
E
B
F
G
C
H
D
25
Do Investors Use Recent Freqs of Reversals?

• Identified teams with same 16 patterns as used by
  Bloomfield and Hales (JFE, 2002) (Table III)
• ?Spread is insignificant for all groups
• Mean changes are not different across low-,
  medium-, high-reversal groups
• Findings are inconsistent with the
  experimental subject results
• Findings are inconsistent with predictions of
  the BSV Model

26
Do Investors Use Recent Freqs of Reversals?

• Sorted observations according to the 256 possible
  8-game historical patterns (Table IV)
• ?Spread is insignificant for all groups
• Mean changes are not different across low-,
  medium-, and high-reversal groups
• Football market participants appear completely
  insensitive to the number of re-
• cent reversals in performance
• Findings are inconsistent
• with BSVs predictions

27
Do Investors Use Longer Histories?

• Expanded histories to include 16- and 30-game
  historical patterns (Table V)
• 8- and 16-game Mean changes are not dif-
  ferent across low- and high-reversal groups
• 30-game Mean changes ARE stat.-signif. differen
  t across low- and high-reversal groups
• Findings for 30-game histories
• are weakly consistent with
• BSVs predictions

28
Do Investors Use Streak Lengths?

• Streak-Based Tests (Table VI)
• Piece-wise regression analysis
• ?Spread a ßHW1HWStrk1 ßHW2HWStrk2
• ßAW1AWStrk1 ßAW2AWStrk2
• ßOpenOpen e, where
• HWStrk1 homes W strk. if homes W strk. lt 3
• 3 if homes W strk. 3
• HWStrk2 0 if homes W strk. lt 3
• homes W strk. 3 if
• homes W strk. 3

29
Spline Transformation of Streak Length
30
Do Investors Use Streak Lengths?

• Streak-Based Tests (contd)
• and AWStrk1 AWStrk2 defined similarly
• Null hypothesis ßi 0 for all i
• Alternative hypothesis ßi gt 0 for i HW1,
  HW2, HL1, HL2 and ßi lt 0 for i AW1, AW2 , AL1,
  AL2

31
Do Investors Use Streak Lengths?

• Alternative hypothesis (predicted by BSV) ßi gt
  0 for i HW1, HW2, HL1, HL2 ßi lt 0 for i
  AW1, AW2, AL1, AL2
• Results ßHW1gt0, ßHW2lt0, ßLW1gt0, ßLW2lt0 all
  stat.-sig.
• Interpretation Bettors expect short streaks to
  continue longer streaks to reverse
• Similar results based on losing streaks

32
Spline Transformation of Streak Length
HLS2
Change in Spread
0.363
HWS1
0.115
0.188
HWS2
HLS1
?0.574
33
Do Investors Use Streak Lengths?

• Alternative hypothesis (predicted by BSV) ßi gt
  0 for i HW1, HW2 ßi lt 0 for i AW1, AW2
• Results ßHW1gt0, ßHW2lt0, ßLW1gt0, ßLW2lt0 all
  stat.-sig.
• Interpretation Bettors expect short streaks to
  continue longer streaks to reverse
• Similar results based on losing streaks

34
CONCLUSIONS

• Performance (against point spreads) is random
• Consistent with Assumption of BSV Model
• Football bettors are relatively insensitive to
  the frequency of recent performance reversals
• Inconsistent w/ Primary Premise of BSV Model
• Bettors expect ? continuations in
  short-run performance ? reversal in performance
  as streak length grows (or exceeds 3)
• Consistent w/ Belief in Regimes, but not as
  hypothesized by BSV

35
CONTACT INFORMATION

• GREG DURHAM
• Assistant Professor of Finance
• Montana State University
• Phone [phone]
• E-mail gregdurham_at_montana.edu