APPENDIX TO MARCH MADNESS, QUANTILE REGRESSION BRACKETOLOGY AND THE HAYEK HYPOTHESIS

Similar documents
Appalachian State University L500030AppStUBlkVinyl. University of Alabama L500030AlabmaBlkVinyl. Arizona State University L500030ArizStBlkVinyl

2010 College Football

Sears Directors' Cup Final Standings


Scoring Algorithm by Schiller Industries

TABLE 3c: Congressional Districts with Number and Percent of Hispanics* Living in Hard-to-Count (HTC) Census Tracts**

2015 State Hospice Report 2013 Medicare Information 1/1/15

DOCTORAL/RESEARCH INSTITUTIONS RECEIVING FULBRIGHT AWARDS FOR

TABLE 3b: Congressional Districts Ranked by Percent of Hispanics* Living in Hard-to- Count (HTC) Census Tracts**

All-Time College Football. Attendance. All-Time NCAA Attendance. Annual Football Bowl Subdivision (FBS) Attendance. Annual Total NCAA Attendance

The American Legion NATIONAL MEMBERSHIP RECORD

U.S. Patents Awarded in 2005 Top 20 Universities

Index of religiosity, by state

TROJAN SEXUAL HEALTH REPORT CARD. The Annual Rankings of Sexual Health Resources at American Colleges and Universities. TrojanBrands.

Unemployment Rate (%) Rank State. Unemployment

Unemployment Rate (%) Rank State. Unemployment

Unemployment Rate (%) Rank State. Unemployment

Unemployment Rate (%) Rank State. Unemployment

Unemployment Rate (%) Rank State. Unemployment

Unemployment Rate (%) Rank State. Unemployment

Unemployment Rate (%) Rank State. Unemployment

Unemployment Rate (%) Rank State. Unemployment

Unemployment Rate (%) Rank State. Unemployment

Unemployment Rate (%) Rank State. Unemployment

Date: 5/25/2012. To: Chuck Wyatt, DCR, Virginia. From: Christos Siderelis

THE METHODIST CHURCH (U.S.)

Table 2 Overall Heterodox-Adjusted Rankings for Ph.D.-Granting Institutions in Economics

PFU DRAFT TIPS Draft Kit. Tip 1: Avoid drafting too many teams from the same conference

Colleges/Universities with Exercise Science/Kinesiology-related Graduate Programs

Drink Mats Grill Mats

5 x 7 Notecards $1.50 with Envelopes - MOQ - 12

MAP 1: Seriously Delinquent Rate by State for Q3, 2008

Interstate Pay Differential

2013 Sexual Health. Report Card. The Annual Rankings of Sexual Health Resources at American Colleges and Universities BRAND CONDOMS

Decline Admission to Boston College Law School Fall 2018

THE GENDER EQUITY SCORECARD VI

Colorado River Basin. Source: U.S. Department of the Interior, Bureau of Reclamation

Name. Class. Year. trojan sexual health report card edition THE ANNUAL RANKING OF SEXUAL HEALTH RESOURCES AT AMERICAN COLLEGES & UNIVERSITIES

Current Medicare Advantage Enrollment Penetration: State and County-Level Tabulations

Engineering bachelor s degrees recovered in 2008

2009 Marketing Academia Labor Market Survey May 20, 2009

ARL ACADEMIC LAW LIBRARY STATISTICS

Rankings of the States 2017 and Estimates of School Statistics 2018

Estimated Economic Impacts of the Small Business Jobs and Tax Relief Act National Report

Keeping Score When It Counts: Academic Progress/Graduation Success Rate Study of 2017 NCAA Division I Men s and Women s Basketball Tournament Teams

2 All-Time College football Attendance. All-Time NCAA Attendance. Annual Football Bowl Subdivision (FBS) Attendance

Critical Access Hospitals and HCAHPS

IU Bloomington Peer Retention & Graduation Rate Comparisons

U.S. Track & Field and Cross Country Coaches Association

STATE INDUSTRY ASSOCIATIONS $ - LISTED NEXT PAGE. TOTAL $ 88,000 * for each contribution of $500 for Board Meeting sponsorship

PLAY TO WIN! Jessica Tidwell, PMP PMI Northern Utah Chapter. All Rights Reserved.

All Approved Insurance Providers All Risk Management Agency Field Offices All Other Interested Parties

Rutgers Revenue Sources

PRESS RELEASE Media Contact: Joseph Stefko, Director of Public Finance, ;

Sentinel Event Data. General Information Copyright, The Joint Commission

Initial (one-time) Membership Fee 10,000 Renewal Fee (every 8 years) $3500

AMERICAN ASSOCIATION FOR AGRICULTURAL EDUCATION FACULTY SALARIES

Voter Registration and Absentee Ballot Deadlines by State 2018 General Election: Tuesday, November 6. Saturday, Oct 27 (postal ballot)

Table 8 Online and Telephone Medicaid Applications for Children, Pregnant Women, Parents, and Expansion Adults, January 2017

EXHIBIT A. List of Public Entities Participating in FEDES Project

Sentinel Event Data. General Information Q Copyright, The Joint Commission

HOME HEALTH AIDE TRAINING REQUIREMENTS, DECEMBER 2016

By Brian L. Yoder, Ph.D.

Weights and Measures Training Registration

Cooperative Program Allocation Budget Receipts Southern Baptist Convention Executive Committee August 2015

Cooperative Program Allocation Budget Receipts Southern Baptist Convention Executive Committee March 2018

2016 INCOME EARNED BY STATE INFORMATION

Mike DeSimone's 2006 College Football Division I-A Top 119 Ratings Bowl Schedule

U.S. Psychology. Departments

Cooperative Program Allocation Budget Receipts Southern Baptist Convention Executive Committee January 2014

Cooperative Program Allocation Budget Receipts Southern Baptist Convention Executive Committee April 2015

Cooperative Program Allocation Budget Receipts Southern Baptist Convention Executive Committee March 2015

Cooperative Program Allocation Budget Receipts Southern Baptist Convention Executive Committee May 2016

Cooperative Program Allocation Budget Receipts Southern Baptist Convention Executive Committee December 2015

Table 6 Medicaid Eligibility Systems for Children, Pregnant Women, Parents, and Expansion Adults, January Share of Determinations

PFU DRAFT TIPS Draft Kit. Tip 1: Avoid drafting too many teams from the same conference

The Regional Economic Outlook

CRMRI White Paper #3 August 2017 State Refugee Services Indicators of Integration: How are the states doing?

Child & Adult Care Food Program: Participation Trends 2017

YOUTH MENTAL HEALTH IS WORSENING AND ACCESS TO CARE IS LIMITED THERE IS A SHORTAGE OF PROVIDERS HEALTHCARE REFORM IS HELPING

Ethnic Studies Asst 55, ,755-2, ,111 4,111

2013 U. of Iowa 86% 85% 87% 2014 U. of Colorado Boulder 84% 86% 86% U. of Nebraska Lincoln 84% 83% 82%

Percentage of Enrolled Students by Program Type, 2016

Acm762 AG U.S. VITAL STATISTICS BY SECTION, 2017 Page 1

2016 NCSEA Structural Engineering Curriculum Survey

FEDERAL R&D FUNDING BY STATE

Statutory change to name availability standard. Jurisdiction. Date: April 8, [Statutory change to name availability standard] [April 8, 2015]

Is this consistent with other jurisdictions or do you allow some mechanism to reinstate?

ARL ACADEMIC HEALTH SCIENCES LIBRARY STATISTICS

States Ranked by Annual Nonagricultural Employment Change October 2017, Seasonally Adjusted

Employment Outcomes, New York / Metro NYC Law Schools

2018 NCAA Division I Men's Basketball Championship

US News and World Report Rankings Graduate Economics Programs Ranked in 2001

Acm769 AG U.S. WATER BAPTISMS, 2017¹ Page 1

UNOFFICIAL. Presentation Score. Cost Score. Penalty

Child & Adult Care Food Program: Participation Trends 2016

HIGH SCHOOL ATHLETICS PARTICIPATION SURVEY

Name: Date: Albany: Jefferson City: Annapolis: Juneau: Atlanta: Lansing: Augusta: Lincoln: Austin: Little Rock: Baton Rouge: Madison: Bismarck:

Keeping Score When It Counts: Graduation Success and Academic Progress Rates for the 2011 NCAA Division I Men s Basketball Tournament Teams

April 17, 2017 Howard Hughes Medical Institute Page 1 of General Investigator Competition List of Eligible Institutions

Transcription:

APPENDIX TO MARCH MADNESS, QUANTILE REGRESSION BRACKETOLOGY AND THE HAYEK HYPOTHESIS ROGER KOENKER AND GILBERT W. BASSETT JR. Abstract. A quantile regression variant of the classical paired comparison model of mean ratings is proposed. The model is estimated using data for the regular 2004-05 U.S. college basketball season, and evaluated based on predictive performance for the 2005 NCAA basketball tournament. Rather than basing predictions entirely on conditional mean estimates produced by classical least-squares paired comparison methods, the proposed methods produce predictive densities that can be used to evaluate point-spread and over/under gambling opportunities. Mildly favorable betting opportunities are revealed. More generally, the proposed methods offer a flexible approach to conditional density forecasting for a broad class of applications. 1. Predictive Densities for Scoring Totals The predictive densities for total scores appear in Figures 1 to 3. The shaded region of the density represents the models preferred bet shaded to the right that the score will be over, and shaded to the left that the score will be under the specified total. As for the point spread figures, we have indicated the estimated probability in the upper left corner of each plot color coded black for successful bets, black for unsuccessful ones. The vertical black line indicates the realized total score, so when it falls in the shaded region of the density the bet is successful and when not, not. 2. Ratings Table 1 reports estimated offensive and defensive ratings of the 2005 NCAA tournament teams based on the fitted quantile regression paired comparison model. Ratings based exclusively on the median (absolute error) model are reported in the first three columns, and mean ratings obtained by averaging over the estimated quantile ratings with respect to Lebesgue measure are reported in the last three columns. Teams are ordered according to the mean ratings. The columns headed Total represent simply the sum of the estimated offensive and defensive ratings and can be interpreted as the slightly bizarre outcome of hypothetical games in which teams play themselves. Figure 4 displays a concise version of the full estimated quantile ratings for the tournament teams. Note that these teams constitute only about one quarter of the teams for which we estimated ratings for the regular season. The scale of the plot, modeled after Tufte s (2006) sparklines, makes it difficult to discern too much in the way of fine detail, but it is evident that some teams are stronger defensively, others Version: April 16, 2008. 1

2 March Madness N CAROLINA ST vs N CAROLINA CHR MISSISSIPPI ST vs STANFORD OLD DOMINION vs MICHIGAN ST TEXAS EL PASO vs UTAH U 0.573 0.495 0.415 0.358 IOWA ST vs MINNESOTA U NEW MEXICO U vs VILLANOVA OHIO UNIV vs FLORIDA U N IOWA vs WISCONSIN U 0.443 0.532 0.482 0.593 UL LAFAYETTE vs LOUISVILLE UCLA vs TEXAS TECH CREIGHTON vs W VIRGINIA TEN CHATANOOGA vs WAKE FOREST 0.43 0.632 0.389 0.442 ST MARYS CALIF vs S ILLINOIS MONTANA U vs WASHINGTON U PITTSBURGH vs PACIFIC U GEO WASHINGTON vs GEORGIA TECH 0.312 0.505 0.639 0.444 Figure 1. Predictive Densities for Total Score of 2005 NCAA Tournament Games

Roger Koenker and Gib Bassett 3 VILLANOVA vs FLORIDA U N CAROLINA ST vs CONNECTICUT S ILLINOIS vs OKLAHOMA ST MISSISSIPPI ST vs DUKE 0.564 0.484 0.49 0.398 ALABAMA BIRMHM vs ARIZONA U W VIRGINIA vs WAKE FOREST NEVADA RENO vs ILLINOIS U GEORGIA TECH vs LOUISVILLE 0.478 0.473 0.502 0.439 CINCINNATI vs KENTUCKY U WISC MILWAUKEE vs BOSTON COLLEGE UTAH U vs OKLAHOMA U PACIFIC U vs WASHINGTON U 0.476 0.345 0.335 0.514 NIAGARA vs OKLAHOMA U IOWA U vs CINCINNATI E KENTUCKY vs KENTUCKY U TEXAS TECH vs GONZAGA 0.461 0.618 0.507 0.658 Figure 2. Predictive Densities for Total Score of 2005 NCAA Tournament Games

4 March Madness WISCONSIN U vs N CAROLINA U LOUISVILLE vs ILLINOIS U MICHIGAN ST vs N CAROLINA U N CAROLINA U vs ILLINOIS U 0.572 0.482 0.538 0.551 VILLANOVA vs N CAROLINA U ARIZONA U vs ILLINOIS U W VIRGINIA vs LOUISVILLE MICHIGAN ST vs KENTUCKY U 0.524 0.569 0.406 0.481 WISC MILWAUKEE vs ILLINOIS U N CAROLINA ST vs WISCONSIN U MICHIGAN ST vs DUKE UTAH U vs KENTUCKY U 0.471 0.508 0.489 0.328 IOWA ST vs N CAROLINA U ARIZONA U vs OKLAHOMA ST W VIRGINIA vs TEXAS TECH LOUISVILLE vs WASHINGTON U 0.461 0.594 0.455 0.532 Figure 3. Predictive Densities for Total Score of 2005 NCAA Tournament Games

Roger Koenker and Gib Bassett 5 stronger offensively and balance is highly desirable. The mean ratings reported in Table 1 are simply the area under the quantile rating curves in Figure 4. University of Illinois at Urbana-Champaign University of Illinois at Chicago

6 March Madness Quantile Ratings for 2005 NCAA Tournament Teams Teams Offense Defense Total N CAROLINA U ILLINOIS U DUKE MICHIGAN ST LOUISVILLE KANSAS U OKLAHOMA ST WAKE FOREST FLORIDA U VILLANOVA OKLAHOMA U CONNECTICUT WASHINGTON U KENTUCKY U ALABAMA U WISCONSIN U GEORGIA TECH ARIZONA U SYRACUSE CINCINNATI PITTSBURGH UTAH U N CAROLINA ST IOWA U BOSTON COLLEGE UTAH STATE MINNESOTA U TEXAS U TEXAS TECH NEW MEXICO U GONZAGA MISSISSIPPI ST LOUISIANA ST W VIRGINIA GEO WASHINGTON IOWA ST S ILLINOIS N CAROLINA CHR PACIFIC U CREIGHTON STANFORD ST MARYS CALIF UCLA ALABAMA BIRMHM WISC MILWAUKEE NEVADA RENO TEXAS EL PASO N IOWA UL LAFAYETTE NIAGARA PENNSYLVANIA OLD DOMINION OHIO UNIV TEN CHATANOOGA Figure 4. Estimated Quantile Ratings Functions for the Teams of the 2005 NCAA Tournament: The vertical axis in each plot represents the unit interval of quantiles, the horizontal axes are uniformly scaled to the range of the estimated coefficients. Offensive ratings, defensive ratings and their sums are plotted. The shaded area gives a visual representation of the mean rating for each team, and teams are ordered by the mean of the offensive plus defensive ratings. These estimates are based on the pre-tournament, regular season games only.

Roger Koenker and Gib Bassett 7 Teams Median Ratings Mean Ratings Offense Defense Total Offense Defense Total N CAROLINA U 75.3-10.7 64.6 75.9-11.3 64.6 ILLINOIS U 67.4-4.1 63.3 65.3-2.1 63.2 DUKE 66.0-4.3 61.7 66.0-6.3 59.7 MICHIGAN ST 67.5-7.7 59.8 64.9-6.1 58.8 LOUISVILLE 66.2-8.5 57.7 66.3-7.9 58.4 KANSAS U 63.5-7.3 56.2 65.2-7.0 58.2 OKLAHOMA ST 63.7-8.4 55.3 65.4-7.7 57.7 WAKE FOREST 72.3-14.9 57.4 72.2-14.5 57.7 FLORIDA U 62.5-5.6 56.9 62.9-6.8 56.1 VILLANOVA 63.5-7.9 55.6 63.3-7.6 55.7 OKLAHOMA U 58.9-8.4 50.5 60.7-5.3 55.4 CONNECTICUT 65.7-11.5 54.2 65.1-9.8 55.3 WASHINGTON U 71.6-16.1 55.5 72.1-17.3 54.8 KENTUCKY U 60.1-7.2 53.0 60.4-5.7 54.8 ALABAMA U 59.5-7.4 52.0 62.4-8.3 54.1 WISCONSIN U 55.6-2.5 53.0 56.5-2.4 54.1 GEORGIA TECH 60.8-7.2 53.5 61.3-7.8 53.5 ARIZONA U 67.9-13.3 54.6 65.8-12.9 52.9 SYRACUSE 61.6-7.2 54.4 61.9-9.0 52.9 CINCINNATI 61.3-7.6 53.6 62.3-9.3 52.9 PITTSBURGH 60.8-7.4 53.4 60.6-7.8 52.8 UTAH U 51.8 0.8 52.6 51.3 1.2 52.5 N CAROLINA ST 60.2-7.2 53.0 58.2-6.2 51.9 IOWA U 62.4-7.5 54.8 62.7-11.1 51.6 BOSTON COLLEGE 59.7-9.3 50.5 58.5-7.1 51.4 UTAH STATE 55.0-6.2 48.8 56.9-5.9 51.0 MINNESOTA U 54.3-8.2 46.2 56.6-5.8 50.8 TEXAS U 61.7-13.2 48.5 61.4-10.7 50.8 TEXAS TECH 59.8-15.1 44.7 62.4-12.2 50.2 NEW MEXICO U 61.9-13.1 48.7 60.4-10.5 49.9 GONZAGA 67.8-13.2 54.5 65.1-15.5 49.6 MISSISSIPPI ST 58.3-8.9 49.4 58.3-9.6 48.7 LOUISIANA ST 60.8-10.5 50.3 63.0-14.7 48.3 W VIRGINIA 56.1-8.2 47.9 56.9-8.7 48.2 GEO WASHINGTON 60.9-15.0 45.9 61.6-13.6 48.0 IOWA ST 55.5-7.4 48.1 55.2-7.4 47.8 S ILLINOIS 51.8-4.1 47.7 52.2-5.3 46.9 N CAROLINA CHR 65.6-19.5 46.1 66.2-19.5 46.7 PACIFIC U 53.9-11.0 42.9 57.0-10.9 46.1 CREIGHTON 56.5-12.0 44.5 57.0-11.1 45.9 STANFORD 56.8-10.5 46.3 56.6-11.0 45.6 ST MARYS CALIF 54.7-9.4 45.3 54.9-9.3 45.6 UCLA 59.7-14.3 45.4 60.2-14.8 45.4 ALABAMA BIRMHM 61.4-14.5 46.9 61.5-16.2 45.4 WISC MILWAUKEE 56.6-14.5 42.1 57.9-12.8 45.1 NEVADA RENO 54.5-8.0 46.5 53.1-8.5 44.6 TEXAS EL PASO 56.8-15.3 41.6 58.3-14.0 44.2 N IOWA 55.8-12.4 43.3 56.5-12.9 43.7 UL LAFAYETTE 58.5-14.0 44.6 56.5-13.9 42.6 NIAGARA 64.1-22.1 41.9 67.7-25.7 42.0 PENNSYLVANIA 50.9-11.2 39.7 52.8-11.0 41.8 OLD DOMINION 51.5-12.2 39.3 52.1-10.7 41.3 OHIO UNIV 54.8-15.1 39.7 56.2-15.5 40.7 TEN CHATANOOGA 50.2-13.1 37.1 50.1-15.3 34.8 E KENTUCKY 48.7-16.4 32.3 49.2-17.1 32.1 MONTANA U 53.0-20.2 32.8 49.4-18.6 30.8 Table 1. Mean and Median Team Ratings for the 2005 NCAA Tournament: As estimated by the quantile regression paired comparison model. Teams are ordered by their mean rating.