Sie sind auf Seite 1von 4

tsset t ** Making crime stationary scatter crime t **So an Augmented Dickey-Fuller test with trend is required gen dcrime

= crime - crime[_n-1] gen crime_lag1 = crime[_n-1] reg dcrime t crime_lag1 predict e, residuals ac e drop e ** Since there is no autocorrelation in the residuals, we include no lags in the test dfuller crime, trend lags(0) ** Since p-value is 0.7801 > 5%, we do not have enough evidence to say that crim e is stationary scatter dcrime t **So an Augmented Dickey-Fuller test with drift is required gen ddcrime = dcrime - dcrime[_n-1] gen dcrime_lag1 = dcrime[_n-1] reg ddcrime dcrime_lag1 predict e, residuals ac e drop e ** Since there is no autocorrelation in the residuals, we include no lags in the test dfuller crime_diff, drift lags(0) ** Since p-value is 0 < 5%, we have evidence that the series is now stationary scatter population t ** Population seems trend stationary scatter cases t **So an Augmented Dickey-Fuller test with drift is required gen dcases = cases - cases[_n-1] gen cases_lag1 = cases[_n-1] reg dcases cases_lag1 predict e, residuals ac e drop e ** Since there is no autocorrelation in the residuals, we include no lags in the test dfuller cases, drift lags(0) ** Since p-value is 0 < 5%, we have evidence that the series is stationary ** So Cases is stationary ** Making unemployment stationary scatter unemployment t **So an Augmented Dickey-Fuller test with trend is required gen unemployment_lag1 = unemployment[_n-1] gen dunemployment = unemployment - unemployment[_n-1] reg dunemployment unemployment_lag1 t predict e, residuals ac e drop e ** Since there is no autocorrelation in the residuals, we include no lags in the test

dfuller unemployment, trend lags(0) ** Since p-value is 0.8331 > 5%, we do not have enough evidence to say that unem ployment is stationary scatter dunemployment t **So an Augmented Dickey-Fuller test with no drift or trend is required gen ddunemployment = dunemployment - dunemployment[_n-1] gen dunemployment_lag1 = dunemployment[_n-1] reg ddunemployment dunemployment_lag1, noconstant predict e, residuals ac e drop e ** Since there is no autocorrelation in the residuals, we include no lags in the test dfuller dunemployment, noconstant lags(0) ** Since test stat is -4.485 < -3, we have evidence that the series is now stati onary ** Making migrationout stationary scatter migrationout t **So an Augmented Dickey-Fuller test with trend is required gen dmigrationout = migrationout - migrationout[_n-1] gen migrationout_lag1 = migrationout[_n-1] reg dmigrationout migrationout_lag1 t predict e, residuals ac e drop e ** Since there is no autocorrelation in the residuals, we include no lags in the test dfuller migrationout, trend lags(0) ** Since p-value is 0.9883 > 5%, we have evidence that the series is nonstationa ry scatter dmigrationout t **So an Augmented Dickey-Fuller test with trend is required gen ddmigrationout = dmigrationout - dmigrationout[_n-1] gen dmigrationout_lag1 = dmigrationout[_n-1] reg ddmigrationout dmigrationout_lag1 t predict e, residuals ac e drop e ** Since there is no autocorrelation in the residuals, we include no lags in the test dfuller dmigrationout, trend lags(0) ** Since p-value is 0.0353 < 5%, we have evidence that the series is now station ary scatter litracy t ** Litracy seems trend stationary scatter gdppercapita t **So an Augmented Dickey-Fuller test with trend is required gen dgdppercapita = gdppercapita - gdppercapita[_n-1] gen gdppercapita_lag1 = gdppercapita[_n-1] reg dgdppercapita gdppercapita_lag1 t predict e, residuals ac e drop e ** Because of autocorrelation in residuals, 0 lags would not work

gen dgdppercapita_lag1 = dgdppercapita[_n-1] reg dgdppercapita gdppercapita_lag1 t dgdppercapita_lag1 predict e, residuals ac e drop e ** Since there is no autocorrelation in the residuals, we include 1 lag in the t est dfuller gdppercapita, trend lags(1) ** Since p-value is 0.5334 > 5%, we have evidence that the series is nonstationa ry scatter dgdppercapita t **So an Augmented Dickey-Fuller test with drift is required gen ddgdppercapita = dgdppercapita - dgdppercapita[_n-1] reg ddgdppercapita dgdppercapita_lag1 predict e, residuals ac e drop e ** Since there is no autocorrelation in the residuals, we include 0 lags in the test dfuller dgdppercapita, trend lags(0) ** Since p-value is 0.2157 > 5%, we have evidence that the series is nonstationa ry scatter ddgdppercapita t **So an Augmented Dickey-Fuller test with no constant is required gen dddgdppercapita = ddgdppercapita - ddgdppercapita[_n-1] gen ddgdppercapita_lag1 = ddgdppercapita[_n-1] reg dddgdppercapita ddgdppercapita_lag1, noconstant predict e, residuals ac e drop e ** Since there is no autocorrelation in the residuals, we include 0 lags in the test dfuller ddgdppercapita, trend lags(0) ** Since test stat is -5.007 < -1.950 , we have evidence that the series is now stationary ** Model 1 ** We want to see that changes in crime in the country are driven mostly by exte rnal conflicts than internal conflicts ** Hence we extimate the following equation: reg dcrime externalconflict internalconflict cases ** Cases has been used as a control for law and order situation predict e, residuals ac e drop e ** We have detected autocorrelation in the residuals of order one ** Hence, we include one lagged term of the dependent variable to cure this ** The resultant model is the following autoregressive equation reg dcrime dcrime_lag1 externalconflict internalconflict cases estat hettest ** Hence, there is no heteroskedasticity estat ovtest ** Hence, there is no functional form misspecification estat durbinalt, robust ** Hence, there is no autocorrelation in the error terms estat archlm ** Hence, there are no arch effects ** Note how external conflict has a positive coefficient, significant at 5% whil

e internal conflict has a statistically insignificant impact on crime test externalconflict - internalconflict = 0 ** This test further yields the result that the coefficient on both the variable s differ statistically. ** Model 2 ** We want to see what affects crime more, ethnic tensions or religious ones ** We set up the following multiple linear regression where duemployment litracy and externalconflict act as controls to limit omitted variable bias reg dcrime ethnictensions religioustensions dunemployment litracy externalconfli ct predict e, residuals ac e pac e drop e ** The two graphs suggests presence of autocorrelation and partial autocorrelati on ** To cure this, we estimate the following model arima dcrime ethnictensions religioustensions dunemployment litracy externalconf lict, ar(1) ma(1) ** Note that the coefficient on religioustensions is bigger in absolute terms. W e test it using the following: test (ethnictensions == religioustensions) ** It yields the result that the two coefficients aren't equal. Hence, religious tensions has a bigger effect on crime than does ethnictensions statistically spe aking.

Das könnte Ihnen auch gefallen