Скачать презентацию A Time Series Analysis of Department of Justice Скачать презентацию A Time Series Analysis of Department of Justice

8e650502edb5f235aa629bfc42ad3ec3.ppt

  • Количество слайдов: 42

A Time Series Analysis of Department of Justice Antitrust Filings: Partisan Politics versus Public A Time Series Analysis of Department of Justice Antitrust Filings: Partisan Politics versus Public Choice Theory Tom Fomby and Dan Slottje Department of Economics SMU

OUTLINE I. Posner’s Seminal (1970) JLE Paper II. The DOJ Count Data III. Using OUTLINE I. Posner’s Seminal (1970) JLE Paper II. The DOJ Count Data III. Using Count Models: Poisson Regression and QML - Negative Binomial Regression IV. The Core Equations to adjust for trend autocorrelation V. Tests of Over/Under-Dispersion VI. Empirical Analysis of Counts VII. A Retrospective View of Posner (1970) VIII. Conclusions

Posner’s (1970) Paper “A Statistical Study of Antitrust Enforcement” JLE • Annual 1890 – Posner’s (1970) Paper “A Statistical Study of Antitrust Enforcement” JLE • Annual 1890 – 1969 • Hypotheses i. Size of Economy (GNP) (+) ii. Size of Budget of Agency (+) iii. Economic Contractions: Scapegoat Hypothesis (Monopoly Causes Contractions) (+) iv. Periods of War: Antitrust could be divisive (-) v. Politics: Party in White House ( - Republicans, + Democrats) vi. Four Years Following Switch of Presidential Party (? ) vi. Presidential Election Year ( ? ) vii. Interactions between Economic and Political Factors • All Informal Findings Were Negative (no association)

Shortcomings of Posner Paper • Relatively Short Data Span. Antitrust Policy has continued for Shortcomings of Posner Paper • Relatively Short Data Span. Antitrust Policy has continued for 34 years hence. • Statistical Analysis Very Informal. For example, he compared simple proportions heuristically and used no formal statistical tests. No analysis of trend autocorrelation in data. • Possibly the use of additional data and more sophisticated statistical methods could shed additional light on factors affecting antitrust activity of DOJ.

DOJ DATA • TOTAL FILINGS = CRIMINAL + CIVIL • ANNUAL: 1891 - 2002 DOJ DATA • TOTAL FILINGS = CRIMINAL + CIVIL • ANNUAL: 1891 - 2002 • SERIOUS FUNDING OF DOJ DID NOT BEGIN UNTIL 1925 • DATA SPAN WE CHOOSE TO ANALYZE IS 1925 – 2002

GRAPH: DOJ FILINGS GRAPH: DOJ FILINGS

DEPENDENT VARIABLES (COUNTS) • TOTAL = TOTAL NUMBER OF CASES • CRIM = CRIMINAL DEPENDENT VARIABLES (COUNTS) • TOTAL = TOTAL NUMBER OF CASES • CRIM = CRIMINAL CASES • CIVIL = CIVIL CASES

POLITICAL EXPLANATORY VARIABLES • PARTY = 1 if Republican, 0 if Democrat • ELECTYR POLITICAL EXPLANATORY VARIABLES • PARTY = 1 if Republican, 0 if Democrat • ELECTYR = 1 if Presidential Election. Year, 0 otherwise • SWITCH = 1 for First Four Years after Party Switch, 0 Otherwise

ECONOMIC (PUBLIC CHOICE) EXPLANATORY VARIABLES • DUNEMP = First Difference of Unemployment Rate • ECONOMIC (PUBLIC CHOICE) EXPLANATORY VARIABLES • DUNEMP = First Difference of Unemployment Rate • DINF 96 = Change in Inflation Rate (1996 dollars) • ERI = DUNEMP – DINF 96 • GDOJ 96 = Growth in DOJ budget (1996 dollars) • GGNP 96 = Growth in GNP (1996 dollars) • WAR = 1 for War Year, 0 Otherwise • RECESS = 1 for negative growth year, 0 0 therwise

CORE EQUATIONS FOR TREND AUTOCORRELATION • TOTAL = f(C, TIME 2, LOG(DOJFILE(1)), LOG(DOJFILE(-2))) Q(12) CORE EQUATIONS FOR TREND AUTOCORRELATION • TOTAL = f(C, TIME 2, LOG(DOJFILE(1)), LOG(DOJFILE(-2))) Q(12) = 3. 5232 (P = 0. 991) • CRIM = g(C, TIME, LOG(CRIM(-1))) Q(12) = 9. 7698 (P = 0. 636) • CIVIL = h(C, TIME 2, LOG(CIVIL(-1))) Q(12) = 3. 5116 (P = 0. 991)

TESTS FOR OVER/UNDER-DISPERSION • TOTAL: Cameron and Trivedi (1990) t = 2. 758 (p TESTS FOR OVER/UNDER-DISPERSION • TOTAL: Cameron and Trivedi (1990) t = 2. 758 (p = 0. 0073) Wooldridge (1997) t = 2. 563 (p = 0. 0123) QMLE parameter = 0. 043 • CRIM: Cameron and Trivedi (1990) t = 3. 227 (p = 0. 0018) Wooldridge (1997) t = 0. 901 (p = 0. 3702) QMLE parameter = 0. 074 • CIVIL: Cameron and Trivedi (1990) t = 3. 968 (p = 0. 0002) Wooldridge (1997) t = 3. 083 (p = 0. 0028) QMLE parameter = 0. 081

PARTISAN POLITICS EQUATIONS PARTISAN POLITICS EQUATIONS

PUBLIC CHOICE EQUATIONS PUBLIC CHOICE EQUATIONS

RETROSPECTIVE POSNER EQUATIONS RETROSPECTIVE POSNER EQUATIONS

CONCLUSIONS - I • We analyze Total, Criminal, and Civil Antitrust filings by the CONCLUSIONS - I • We analyze Total, Criminal, and Civil Antitrust filings by the Department of Justice over the years 1925 – 2002. • We find that Partisan Politics (Party) doesn’t seem to affect any of the filings of the DOJ. In other words, when it comes to Antitrust enforcement, Democrats and Republicans are alike in terms of their activism/passivity, other factors held constant.

CONCLUSIONS - II • Turnover in administrations doesn’t appear to bring with it reactionary CONCLUSIONS - II • Turnover in administrations doesn’t appear to bring with it reactionary change vis-à-vis the previous administration. (Switch) • Election year politics (Electyr) doesn’t seem to affect the number of antitrust cases brought by the DOJ. That is, Antitrust activity of the Presidential election year appears to be no different than that of non-election years.

CONCLUSIONS - III • The impact of economic (Public Choice) variables on DOJ Antitrust CONCLUSIONS - III • The impact of economic (Public Choice) variables on DOJ Antitrust activity comes through a select few variables and then only affects Total and Criminal filings and not Civil filings. Possibly Criminal cases have a higher profile than Civil cases and as a result are more important in conveying messages to the Public about the Administration’s concern over economic variables of interest to the public. • Evidently DOJ officials engage in Antitrust activity with a reticence that depends on the most recent changes in the unemployment rate and inflation rate. We measure this reticence with what we call the “Economic Reticence Index” (ERI = dunemp – dinf 96).

CONCLUSIONS - IV • DOJ officials appear to be more reticent in engaging in CONCLUSIONS - IV • DOJ officials appear to be more reticent in engaging in Antitrust activity when unemployment is increasing (possibly in fear of creating more unemployment) and less reticent in bringing Antitrust cases when inflation is increasing (possibly thinking that inflation is being caused by “monopoly power”). • There is tenuous evidence that DOJ officials may weigh unemployment somewhat more heavily than inflation when deciding on the vigor with which to pursue Antitrust activity.

CONCLUSIONS - V • Finally, we examine an interesting historical question. If Prof. (Judge) CONCLUSIONS - V • Finally, we examine an interesting historical question. If Prof. (Judge) Posner had, in 1969, the econometric methodology of today, would he have found the same ERI effect that we found here? Or, even in the presence of advanced econometric methodology, would the limited span of the data he had available at the time have prevented him from finding any meaningful relationships at all as implied by the lack of associations he reported in his 1970 paper? • A related question: Do advanced econometric methods help steepen the learning curve in economics?

CONCLUSIONS - VI • In fact, if Prof. Posner had the current econometric methodology CONCLUSIONS - VI • In fact, if Prof. Posner had the current econometric methodology available at the time he wrote his 1970 paper, he too would have found the ER effect (some 34 years earlier than this paper). To whit, yes, advanced econometric methodology, can help steepen the learning curve of economics.

POSTSCRIPT POSTSCRIPT

COMPETING PARADIGMS • Poisson Assumption: Var(y|x) = E(y|x) (1) • Poisson GLM Variance Assumption: COMPETING PARADIGMS • Poisson Assumption: Var(y|x) = E(y|x) (1) • Poisson GLM Variance Assumption: Var(y|x) = (2) • Negative Binomial Assumption: Var(y|x) = E(y|x) + (3) • Two-Step Negative Binomial QMLE: Replace with

Neither (2) or (3) is more general than the other. In other words (2) Neither (2) or (3) is more general than the other. In other words (2) is not encompassed by (3) or vice versa. It is difficult to compare Poisson QMLE and 2 step NB QMLE on efficiency grounds. Poisson QMLE is “second order efficient” in a neighborhood of. Practically speaking, this means that the Poisson QMLE is almost as asymptotically efficient as the 2 -step NB QMLE for small amounts of over-dispersion, .

A NEW ARTICLE Ghosal, Vivek and Gallo, Joseph (2001) “The Cyclical Behavior of the A NEW ARTICLE Ghosal, Vivek and Gallo, Joseph (2001) “The Cyclical Behavior of the Department of Justice’s Antitrust Enforcement Activity, ” International Journal of Industrial Organization, Vol. 19, 27 – 54. Major Findings: • Antitrust Case Activity is Countercyclical (i. e. Weaker economy implies more cases; stronger economy implies less cases. ) • Number of Cases positively affected by funding • Political Affliation of President and Republican Vs. Democratic Composition of House and Senate Do not have clear impact on case activity

Differences: • Span of Data – Annual 1955 – 1994 • Methodology: Multiple Regression Differences: • Span of Data – Annual 1955 – 1994 • Methodology: Multiple Regression Cases = f(c, t, EAM = Economic Activity Measures = S&P 500, DOW, Profit/Income, Profit/Consumption, GDP, UN POL(1) = Pres x% Republications in House + Senate POL(2) = Pres x% Repub in House x%Repub in Senate