DDCG Notes

Intro

  • Daron Acemoglu

  • James A Robinson

  • Suresh Naidu

  • Pascual Restrepo

published 2019, Journal of Political Economy

Problems for the question:

  • Democary Indices

  • institutional differences between Dem / non-Dem

  • correlation with other changes

  • slumps in GDP before democratization

=>  DiD / panel data estimates not good idea

Literature

Notes here also from other papers

Question dates back to 1959 (lipset hypothesis)

Lipset: \(\text{economic growth} \to \text{democracy}\)

1990s:

  • conitnius variables of democracy + simple regressions (OLS, problematic)
  • (eg [@barroDemocracyGrowth1996])
  • no clear /negative effect
  • confirm Lipset hypothesis (growht => democracy)

Barro: \(\text{democracy} \to \text{growth} \downarrow\)

2000s:

  • binary measures of democracy + Diff in Diff
  • mixed / postiive / not significant effects
  • eg [@giavazziEconomicPoliticalLiberalizations2005]

Giavazzi: \(\text{democracy} \rightsquigarrow \text{growth}\)

modern aproaches: Acemoglu

  • different strategies
  • as guess from Title: positive results

Acemoglu: \(\text{democracy} \implies \text{growth}\)

Meta Analysis [@colagrossiDoesDemocracyCause2020]

Approaches

First:

  • country fixed effects

  • control with lags

    • esp. pre-dem GDP dip

=> ensure that deomcratizations are (conditionally) uncorrelated to past GDP

=> robust estimates of 20% higher GDP pC after 25 years

Second:

  • semiparametric treatment effects framework

    • statistical model to analyze treatment effect on outcome

    • mix of parametric and non-parametric methods

    • flexible approach

  • democratization influences distribution of potential GDP afterwards (time-dsitribution)

Third:

  • Instrumental Variable Approach

  • regional waves of democratization

    • differ from economic shocks

=> all approaches 25% increase

Channels:

  • investment +

  • schooling +

  • economic reforms

  • public services +

  • social unrest -

possible other Literature

  • Schumpeter: theoretical arguments against

  • Barro: empirical arguments against

  • Comments on Paper?

Data

Democracy INdex (consolidated and dichotomous)

Sources:

  • Freedom House (free, partly free, unfree)

  • Polity IV (-10, +10)

  • Cheibub et al (@cheibubDemocracyDictatorshipRevisited2010)

  • Boix et al (@boixCompleteDataSet2013)

measure is also for short lived democracies!

either 0 or 1

Outcome Variable: log GDP p.C 2000 Dollars (World Bank)

Regression

First Approach

\[ y_{ct} = \beta D_t + E_{j=1}^p \gamma_j y_{ct-j} + \alpha_c + \delta_t + \epsilon_{ct} \]

Formula: …

  • \(y_ct\) = log GDP per capita in country c at time t

  • \(D_{ct}\) = Dichotomous measure of democracy

  • p lags of log GDP for control

  • \(\alpha\) = country fixed effects

  • d = time fixed effects

  • \(e\) = error

\[ \frac{ \hat\beta }{1- E_{j=1}^p \hat\gamma_j} \]

lags = countries not on different GDP trends before (stop reverse causality)

Results: (Table 2) weirdly split on two pages

Estimators:

  • within estimator (aka fixed effects model)

  • GMM estimator (@arellanoTestsSpecificationPanel1991)

  • HKK Estimates(@hahnBiasCorrectedInstrumental2001)

why switch?

  • Nickell Bias (not enough time periods compared to entities), always in within estimator

    • just theoretical

    • they have enough times, therefore ok

  • bias of fixed-effects estimator

  • use other methods, but not that widespread

    • GMM has their own biases (asymptitic bias, too many Ts)

    • HHK similar to first

Robustness:

  • omitted variable between GDP and Democracy

    • other approaches

  • different levels of income before democratization

    • solve by using just subsample with similar wealth

  • have dummy for post-soviet transition to democracy

  • other ommited variables controls:

    • unrest before democracy is good for growth

      • also controlled for in other approaches and lags of unrest

    • external trade influences both

      • control for external financial flows

    • demographic changes

      • also controlled for

Second Approach

Prblem of First: linearity assumption of GDP growth

semi-parametric

  • no parametric assumption for GDP development

  • but assumption for likelihood of transition to democracy

Assumption:

  • transitions to democracy preceded by dip

  • no other confounding factors that influence propensity to democratize

Results:

  • confirmation of assumption

  • results similar to first approach

  • used three different methods

Third Approach

waves of democratization

  • external factor that influences both demotracy and GDP

  • alleviates errors in democracy measure

  • regionally limited due to similiar politics  culture etc.

e.g Soviet Union fall

7 regions

  1. Africa

  2. East Asia + Pacifics

  3. Eastern Europe + Central Asia

  4. Western Europe + other developed countries

  5. Latin America + Carribean

  6. MENA

  7. South Asia

Wave definition

  • significant determinants of democracy

  • but wothout trend effect on GDP

=> different Approach, but similiar estimates

Mechanisms

potential channels and data

  • % investment GDP (logs)

  • TFP (log)

  • measure of economic reform (Giuliano 2013) 0-100

  • % trade GDP (log)

  • % taxes of GDP (log)

  • primary school enrollment

  • secondary school enrollment

  • child mortality (log)

  • social unrest dummy

Estimation

\(m_{ct} = \beta D_{ct} + \sum_{j=1}^p \gamma_j y_{ct-j} + \sum_{j=1}^p \eta_j m_{ct-j} + \alpha_c + \delta_t+\epsilon_{ct}\)

Results: Democracy =>

  • economic reforms

  • tax rev.

  • enrollment school

  • some evidence

    • invstment

    • opnennes to trade

    • less social unrest

=> Democracy: more taxes, more investment in school, econ. reforms

Critiqie of DDCG:

  • Dem. needs certain preconditions (human capital, institutions)

  • Team controls for

  • => high human capital (educated), democracy = more growth

    • high ed. = reduce stakes of distributional conflicts

Conclusion

Skepticism about democracy always existed (see Plato)

but effects of democarcy are there!

  • more than others argued (esp Barro)

Critique

  • not enough explaining, why choose which specification

    • in every approach choose the one that is most similar results to others

  • dichotomous democracy index

    • democratization is a flowing element

  • regional analysis

    • size  of regions

Linear Dynamic Panel Model

  • Linear: Assumes proportional relationships between variables.

  • Dynamic: Includes lagged values of the dependent variable to capture temporal dependencies.

  • Panel: Analyzes data with both cross-sectional (different entities) and time-series (observations over time) dimensions.

  • Fixed Effects: Incorporates parameters specific to each entity in the panel, capturing time-invariant characteristics.

  • Model Purpose: Used to analyze how changes in variables, such as GDP, relate to each other over time, considering individual entity characteristics and historical patterns.