Forecasting in Financial and Sports Gambling Markets: Adaptive Drift Modeling

Copyright © 2011 by John Wiley & Sons, Inc. All rights reserved.

Published by John Wiley & Sons, Inc., Hoboken, New Jersey

Published simultaneously in Canada

No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except as permitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization through payment of the appropriate per-copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400, fax (978) 750-4470, or on the web at www.copyright.com. Requests to the Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/permission.

Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives or written sales materials. The advice and strategies contained herein may not be suitable for your situation. You should consult with a professional where appropriate. Neither the publisher nor author shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages.

For general information on our other products and services or for technical support, please contact our Customer Care Department within the United States at (800) 762-2974, outside the United States at (317) 572-3993 or fax (317) 572-4002.

Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic formats. For more information about Wiley products, visit our web site at www.wiley.com.

Library of Congress Cataloging-in-Publication Data:

Mallios, William S.

Forecasting in financial and sports gambling markets: adaptive drift modeling / William S. Mallios.

Includes bibliographical references and index.

ISBN 978-0-470-48452-4

Printed in Singapore

10 9 8 7 6 5 4 3 2 1

Preface

Adam Smith's last words in 1790: I believe we should adjourn this meeting to another place. The meeting is re-adjourned—as it has been so many times. The focus is on adaptive drift modeling—models that adapt to the specific evolving market and then drift over time. Forecasting procedures are for purposes of active trading in finance and betting against the line in sports. Such modeling has relevance to other evolving markets, including the influenza markets.

There is ample motivation for this book: the recent financial market abyss, the proliferation of sports gambling, the metastasis of lotteries, the ongoing epidemics of financial and mathematical illiteracy—epidemics that are allied with the emerging epidemic of adolescent problem gambling—and the assortment of sausage legislation proposed for financial market reform and regulation and legalized online gambling. All of these topics are driven by a common denominator: human behavior—the type of behavior that has been described as the madness of crowds combined with the cunning of the few.

Thomas Huxley, the 19th-century biologist and defender of Darwin, said that the great tragedy of science was the slaying of a beautiful hypothesis by an ugly fact. A counter opinion is that the beauty of science is the eventual slaying of outdated hypotheses and dogma through evolving and enlightened inquiry.

For over 50 years, the prevailing investment wisdom was buy and hold for the long term. Indeed, markets were said to be efficient in which case active or short-term trading would inevitably result in portfolio losses. Researchers were quick to apply efficient market dogma to the sports gambling markets: that is, that over the longer term, the bookmakers' lines can't be beaten. Popular books such as A Random Walk Down Wall Street (Malkiel, 1985) portrayed active traders as inevitable losers. Even the Wall Street Journal carried a series that compared dart throwing in selecting equities with selections by expert traders.

However, with financial market deregulation and the entry of hedge funds, particularly during the latter stages of the Clinton Administration, active trading strategies began to dominate the buy-and-hold strategies. Financial innovations began to sprout under the cover of efficient market dogma. Finally, the innovations bubbled and the dogma crumbled. In testimony before Congress in 2006, former Fed Chairman Alan Greenspan humbly admitted: I made a mistake in presuming that the self interests of organizations, specifically banks and others, were such that they were best capable of protecting their own shareholders and their equity in the firms.

In lay terms, hedge funds are investment vehicles limited by law to the very rich. In contrast to mutual funds, they are largely unregulated and invest opaquely. They hedge their investment monies, not so much in the sense of hedging or protecting against risk, but rather for purposes of maximizing profits. Hedge funds are known as quant funds when they employ quantitative (statistical) modeling in forecasting short-term price movements. Quant funds were subjected to severe criticism when the subprime mortgage crisis spilled over to other financial markets—at which point the hedge funds were affected adversely. In many quarters, quant modeling was condemned, sometimes in buffoonish fashion, along with the efficient market hypothesis.

When you see a quantitative expert, shout for help, call for his disgrace, make him accountable. Ask for the drastic overhaul of business schools.… Ask for the Nobel Prize in Economics to be withdrawn from the authors of these theories, as the Nobel's credibility can be extremely harmful.

(Taleb and Triana, 12/8/08)

A number of explanations were given for the forecasting failures of quant modeling—failures that also apply to forecasting in the sports gambling markets.

1. It is typically the case that modeling complexities induce less capable analysts to impose invalid or oversimplified modeling assumptions, which usually lead to invalid forecasts, especially during periods of unexpected volatility.

2. Quant models have short shelf lives and tend to be of limited value when they are not updated on a continuing basis to accommodate changing market dynamics.

3. Recent relevant and vital information, often from allied sciences, is not incorporated in the model-building procedure. (For example, the weather determines price differences in the agricultural commodities market, which has led hedge funds to hire meteorologists to interact with their traders.)

4. With so few qualified modelers, the best modelers are lured away by competing funds. The modelers then use the same models to chase the same money.

5. Market shocks (i.e., unexpected, often unpredictable events) are either not incorporated or are incorporated inappropriately in forecasting models. Moreover, there has been a failure to recognize that the volatility associated with sufficiently large shocks may destabilize model structure, at least temporarily, to the extent that model forecasts become unreliable.

6. In situation 5, there is typically a failure to reconstruct and adapt the forecasting model so that it applies to evolving market conditions, which includes adapting the model to incremental changes during periods when markets are relatively stable.

Given the highly competitive and risky environments of current-day financial and sports gambling markets, the focus is on the dynamic process of constructing effective forecasting rules that are based on both graphical patterns and adaptive drift modeling of cointegrated time series. The graphical patterns are in terms of candlestick charts and their variants, a well-known charting procedure dating back to feudal Japan. Charting objectives are to identify optimal time periods in financial markets and optimal games in sports gambling markets for which forecasting rules and models are likely to provide profitable trading or wagering outcomes.

The modeling of cointegrated time series means that forecasts are with reference to a system of simultaneous time series wherein long-term relations exist between the individual series comprising the system. Disequilibria between such relations are known to affect subsequent outcomes within individual relations. As such, estimates of the between-series disequilibria can be used in forecasting subsequent movements within the individual time series. For example, consider a time-varying, emotional attachment variable for each of two lovers. The two variables are clearly related, but the relation between the two is subject to disequilibria over time. When, at any point in time, a major disequilibrium occurs—in the sense of, say, a temporarily strained personal relationship—the tendency is for the relation to return subsequently to normal. In this case, between-relation disequilibria can be used to predict subsequent outcomes for each of the individual variables. On the other hand, the disequilibria can become sufficiently large—analogous to periods of extreme volatility in financial markets—to the extent that the lovers may split (temporarily or permanently) and their responses may no longer be cointegrated.

Optimal profit-making situations in financial markets occur when markets are inefficient, in which case short-term price movements are more likely to be predictable. In the sports gambling markets, periods of market inefficiency are in terms of forthcoming games where outcomes are likely to differ considerably from the bookmakers' lines.

Shocks, defined as unexpected deviations from the norm (or from what is expected), may or may not be predictable. However, once they occur and are known or estimated, their effects are often highly consequential in effectively forecasting subsequent outcomes. In fact, shocks are the key to successful forecasting in the markets under study.

Shocks are best illustrated in modeling National Football League or National Basketball Association game outcomes. A bookmaker's line on a game is based on the gambling public's expectation of what the game outcome will be. Specifically, the bookmaker's job is to determine that line (or spread) which evenly divides the money wagered on the game. Since the parties covering the bets charge a commission (usually, 10 percent) on each bet that is made, it is irrelevant whether the line is realistic or not as long as the payouts to the winners are covered by the losers' losses.

To illustrate the effects of gambling shocks, suppose that a heavily favored team is upset by an underdog, such as having the 2008–2009 Los Angeles Lakers, an 11.5-point favorite, lose to the Sacramento Kings in midseason. The likely Laker team reaction to the loss is to reevaluate game strategies, identify mental lapses, elevate testosterone levels, and then make up for the miserable performance not only in their next game or games but also in their next meeting with the Kings. In this context, the gambling shocks are reflections of physiological–psychological–sociological variables that affect player and team personnel. As such, shocks tend to be determining factors in subsequent game outcomes. Discussions of shock effects in financial markets, termed moving average effects, are presented in Chapter 4.

The creation of sports hedge funds appears inevitable—if they do not already exist in the opaque and ill-regulated world of hedge funds. A bet on the favored 2009 New York Yankees in October carried less risk than an active trader's long or short position on Bank of America during the same time period—at least for bettors without access to insider information. In a similar vein, online sports gambling will eventually be legalized for purposes of enriching government coffers—in the same way that Prohibition was repealed to provide lucrative tax revenues. Concurrently, the lottery markets will continue to flourish in the form of stupidity taxes that prey on those who are infected by the raging epidemics of mathematical and financial illiteracy and the related epidemic of adolescent problem gambling.

The great economist Woody Allen once said: More than any time in our history, mankind faces a crossroads. One path leads to despair and utter hopelessness, the other to total extinction. Let us pray we have the wisdom to choose correctly. Financial and sports gambling markets will continue to be an inevitable part of the economic and social fabric for unforeseeable future generations. Reasonable courses must be chartered. This meeting will be readjourned again and again and again.

Updates of adaptive drift modeling forecasts in sports gambling and financial markets are available at www.MalliosAssociates.com.

Acknowledgments

Throughout the writing of this book, Ronna Mallios provided both expertise and critiques. The association between gradual and abrupt drift in modeling and the Darwin–Gould–Eldredge theories resulted from communications with Seth Mallios. Peter Mallios contributed literary criticisms. Bo Hatfield provided the means of converting sports and financial data bases into formats that allowed applications of adaptive drift forecasting.

William S. Mallios

1

Introduction

1.1 Favorable Betting Scenarios

The buy-and-hold strategies under efficient market dogma have shifted toward active trading strategies under adaptive market alternatives. Microeconomics appears to be back. It would be better if, as Keynes said, markets were not the by-product of a casino. But, in fact, they are.

In light of the greatest downturn since the Great Depression, the shift to active trading is not without critics. Under Saint Joan's banner, French President Sarkozy has taken steps to instill moral values in the global market economy by urging policymakers to consider fresh ways of combating financial short-termism¹ (Hall, 1/3/09). Perhaps Mr. Sarkozy has taken a perverse view of Keynes' dictum that economics is a moral and not a natural science.

The recessionary angst of late 2008 saw many favorable betting scenarios in financial and sports gambling markets. Attractive bets included: establishing short positions on Goldman Sachs shares during November and betting on the Los Angeles Lakers (favored by 3 points) in their Christmas Day rematch with the Boston Celtics. (The Celtics embarrassed the Lakers in the previous National Basketball Association championship series). The attractiveness of each bet depended on the effectiveness of the gambler's forecasting models—models that are assumed based on public information.

It has been argued that profitable modeling forecasts tend to favor the sports gambling markets since they are accommodated by greater regulation and surveillance, considerably less opacity, and public point spreads that reflect the gambling public's expectations. For example, the New England Patriots' loss to the New York Giants in the 2008 Super Bowl was an outcome that superseded the New York Jets' upset win over the Baltimore Colts in the 1969 Super Bowl. The Patriots were prohibitive 12-point favorites; the bookmakers' line on total points scored was 53.5. Relative to the lines on the difference and total points scored, the Patriots had vastly overperformed throughout the first half of the season, then underperformed but kept winning until the finale (see Figure 1.2.2). New England had clearly peaked by midseason.

In contrast, the Giants jelled in the second half of the season and peaked during the play-offs (see Figure 1.2.3). In the finale, the Giants won 17–14, an outcome that was easily amenable to effective forecasting; see Table 1.1.1 and the modeling procedure described in Section 10.2.

Table 1.1.1 Super Bowl 2008: NE Patriots vs. NY Giants + 12a

aNYG expected winning margin: 3.4 points. Outcome: NYG won by 3 points.

Relative to the line, the Giants' expected winning margin of 3.4 points was a far more realistic estimate of the outcome. (See Section 11.2 for the calculation of the expected winning margin.) However, whether or not the line is realistic, there are always two groups of winners—those covering² the bets and those betting on the winning side of the line—and one group of losers—those betting on the losing side of the line. Those covering the bets charge a commission per bet and are always the winners as long as the line splits the money wagered (i.e., losing bets pay off the winning bets after commissions). Thus, a bookmaker's line is simply a measure of the gambling public's expectation of a game's outcome—regardless of whether or not that expectation is realistic.

A financial market analogy to Table 1.1.1 is illustrated in terms of Microsoft's (stock symbol: MSFT) price movements during 1999–2000, a volatile period during final inflation and deflation of the NASDAQ bubble. From 3/27/00 to 4/3/00, the MSFT closing price dropped from $53.13/share to $44.53/share. Figure 1.1.1 presents weekly price changes and volumes through the January–June 2000 period, and Figure 1.1.3 presents these changes in terms of a candlestick chart (see Section 5.1 for detailed discussions).

Figure 1.1.1 Line chart of weekly per share closing prices, including weekly volumes, for Microsoft (MSFT).

In Figure 1.1.3, each week in Figure 1.1.1 is represented by a candlestick that depicts four summary prices for MSFT: the opening price (O), the high (H), the low (L), and the closing price (C) for the week. A candlestick is composed of a body and a wick that extends above and below the body. The body is white if C > O and dark if O > C. The maximum (minimum) of the wick is the high (low) for the week; see Figure 1.1.2 for an illustration of three hypothetical candlesticks. The lower portion of Figure 1.1.3 presents the 25- and 100-day moving averages for C (where five trading days correspond to one week). The moving averages are based on successive days prior to each weekly candlestick.

Figure 1.1.2 Three candlesticks with (1) H > O > C > L, (2) H > C > O > L, and (3) H > O = C > L.

Figure 1.1.3 Candlestick chart for weekly per share Microsoft price changes including 25- and 100-day moving averages and weekly volumes.

(Source: MSN Money)

Figure 1.2.1 2008–2009 NBA play-off games for Los Angeles Lakers with wins over Utah (5 games), Houston (7 games), Denver (6 games), and Oriando (won NBA title in 5 games).

Figure 1.2.2 New England Patriots: 2007–2008 regular season and three play-off games, ending with Super Bowl loss to New York Giants.

Figure 1.2.3 New York Giants: 2007–2008 regular season and four play-off games, ending with Super Bowl win over the New England Patriots.

A short-term modeling objective was to forecast the change in the closing price from 3/27/00 to 4/3/00; see box 1 in Figure 1.1.1. Adaptive drift modeling led to the results in Table 1.1.2 (see Chapter 9). The forecast correctly projected a significant drop in price, although the actual loss was underestimated relative to the expected loss. The same modeling procedures were used to forecast losses correctly through mid-May and gains in the rebound that followed; see box 2 in Figure 1.1.1.

Table 1.1.2 Odds on D(MSFT, t), the Change in the Closing Price per Share of Microsoft from 3/27/00 to 4/5/00a

aExpected gain/loss for D(MSFT, t) = − $4.24. Observed gain/loss for D(MSFT, t) = − $9.60.

1.2 Gambling Shocks

A gambling shock (GS) is defined as the difference between the game outcome and the line. For example, if the line on the difference favors the Patriots by 12 points and they lose by 3 points, GSdifference(NE) = GSD(NE) = − 3 − (12) = − 15. If the line on the total points scored in the Giants–Patriots game is 51 and the total points scored is 31, then GStotal(NE) = GStotal(NYG) = GST( * ) = 31 − 51 = − 20. Larger values of |GSD| and/or |GST| for a particular team generally affect that team's subsequent performance or performances in that they may reflect the effects of motivation, injuries, personnel problems, and so on—all of which translate into physiological, psychological, and sociological variables.

When, for example, the Giants suffered through two embarrassing losses to the Dallas Cowboys during the 2007–2008 regular season, the likelihood of a Giants' upset win against Dallas in the play-offs was exceptionally high (especially in view of the Giants late-season performances). In fact, when the Giants lost a game throughout the regular season, they usually won their next game (as shown in Figure 1.2.3).

When there are marked differences in player talent between opposing teams, the GS may act as a surrogate for fans and teams in the evaluation of team and player performances. The home team fans may take consolation when their underdog team loses by less than the spread—especially if they've bet on their team.

We play hard and cover. We lead the league in covering the point spread.

(Hubie Brown, coach of the last-place New York Knicks, Sports Illustrated, 1986)

Figure 1.2.1 depicts game outcomes and accompanying gambling shocks for the 23 Los Angeles Lakers' play-off games leading to their 2008–2009 National Basketball Association (NBA) title. The Lakers won in five games against Utah, seven against Houston, six against Denver, and then five in the finale against Orlando. White bodies denote games in which the Lakers beat the line on the difference. The minimum value of the white body is the line on the difference for Lakers, and the maximum of a white body is the Lakers' winning/losing margin. Dark bodies denote games in which the Lakers did not beat the line; that is, the maximum value of a dark body is the line on the difference, and the minimum value of a dark body is the winning or losing margin. A white (dark) body indicates that the Lakers overperformed (underperformed) relative to the line on the difference. The size (magnitude) of the body reflects (equals) the size of the gambling shock on the difference for the Lakers. An observed difference above (below) zero signifies a Lakers' win (loss).

The size of the gambling shock for the total points scored is given by the wick (or stick) that extends either above or below each body. When the wick extends above (below) the body, GStotal > 0  (GStotal < 0). For example, in the first play-off game against the Utah Jazz, the Lakers were favored by 12 points and won 113–100; the line on the total was 210.5. Thus, GSdifference(LAL) = 13 − 12 = 1 (a small white body) and GStotal(LAL) = 213 − 210.5 = 2.5 (a short wick extending above the white body). In this game, the gambling public's