Dynamic Panel Data (Arellano-Bond)

The Arellano-Bond estimator is a generalized method of moments (GMM) estimator for dynamic panel data models that include a lagged dependent variable, using deeper lags as instruments in first differences.

Arellano-Bond 추정량은 시차 종속변수를 포함하는 동적 패널모형에서 1차 차분 후 더 깊은 시차를 도구변수로 사용하는 일반화적률법(GMM) 추정량이다.

How do I estimate a panel regression that includes the lagged outcome as a predictor?

결과변수의 시차값을 설명변수로 넣는 패널 회귀는 어떻게 추정하나요?

The Arellano-Bond difference GMM estimator removes unit fixed effects by first-differencing and instruments the lagged outcome with earlier lags, avoiding the bias that fixed effects estimation has with a lagged dependent variable in short panels. It fits panels with many units and few time periods where the outcome depends on its own past value.

When to use Dynamic Panel Data (Arellano-Bond)

• Panels with many units and relatively few time periods
• The outcome depends on its own previous value, such as persistence in firm investment or growth
• Some regressors are endogenous or predetermined and need internal instruments

When to use something else

• The panel has many time periods: the fixed effects bias is small, so a fixed effects model may suffice
• The outcome is highly persistent so lagged levels are weak instruments: use system GMM (Blundell-Bond)

Assumptions and requirements

• No second-order serial correlation in the differenced errors
• Instruments are valid (exogenous), checked with the Hansen or Sargan test
• Number of instruments kept small relative to the number of units
• Many units (large N) relative to time periods (small T)

What the analysis reports

• Coefficients including the lagged dependent variable with robust standard errors
• Arellano-Bond tests for AR(1) and AR(2) in differenced residuals
• Hansen or Sargan test of overidentifying restrictions
• Number of instruments and number of groups

How to report it

Difference GMM estimates showed a persistence coefficient of [value] (SE = [value]); the AR(2) test (p = [value]) and Hansen test (p = [value]) did not reject instrument validity.

Related methods

• System GMM (Blundell-Bond)
• Panel Fixed Effects Model
• Instrumental Variables Regression

Run Dynamic Panel Data (Arellano-Bond) on Quorant