Abstract
In this paper, we suggest a J test in a dynamic spatial panel framework of a null model against one or more alternatives. The null model we consider has fixed effects, along with nonparametrically specified spatial and time dependence. The alternatives can have either fixed or random effects with the same complications. We implement our procedure to test the specifications of a demand for cigarette model. We find that the most appropriate specification is one that contains the average price of cigarettes in neighboring states, as well as the spatial lag of the dependent variable. Along with formal large sample results, we also give small sample Monte Carlo results. Our large sample results are based on the assumption \(N\rightarrow \infty \) and T is fixed. Our Monte Carlo results suggest that our proposed J test has good power and proper size even for small to moderately sized samples.
Similar content being viewed by others
Notes
There is, of course, a large literature relating to the J test. For example, see Davidson and MacKinnon (1981), MacKinnon et al. (1983), Godfrey (1983), Pesaran and Deaton (1978), Dastoor (1983), Pesaran (1974, 1982), Delgado and Stengos (1994) and the reviews given in Greene (2003, pp.153–155, 178–180) and Kmenta (1986, pp. 593–600). A nice overview of issues relating to non-nested models is given in Pesaran and Weeks (2001).
See, e.g., Anselin et al. (2008), Kapoor et al. (2007), Baltagi et al. (2003, 2007a, c, 2013), Baltagi and Liu (2008), Debarsy and Ertur (2010), Elhorst (2003, 2008, 2009, 2010), Elhorst and Freret (2009), Elhorst et al. (2010), Lee and Yu (2010a, b, c, d), Mutl and Pfaffermayr (2011), Pesaran and Tosetti (2011), Yu and Lee (2010), Yu et al. (2008), Parent and LeSage (2010), Piras (2013).
In a slightly different context, Mutl and Pfaffermayr (2011) suggest and give large sample results for a Hausman test to discriminate between such models.
This was pointed out by a referee.
For the estimation of a dynamic panel data model with random effect, see Parent and LeSage (2012).
Clearly if the model also had regressors which only varied cross-sectionally, the corresponding coefficients would not be identified. This was the case in a paper by Kelejian et al. (2013).
See, e.g., Kapoor et al. (2007).
As a simple illustration, note from (1), (3), and (4) that
$$\begin{aligned} u_{T}= & {} R_{T1}\varepsilon _{1}+R_{T2}\varepsilon _{2}+\cdots +R_{TT}\varepsilon _{T} \\ u_{1}= & {} R_{11}\varepsilon _{1} \end{aligned}$$Thus, since \(y_{1}\) depends on \(u_{1},\) which in turn depends on \( \varepsilon _{1}\), as does \(u_{T},\) \(y_{1}\) can not be viewed as exogenous even at time period T.
Essentially, the rational of the J test is that if the null model is correct, these augmenting variables should not add to the explanation of the dependent variable in (9).
The proof for these statements is tedious. It can be obtained by writing to the authors.
Further specifications are given below.
We indicate that the value of i should be preselected in order to avoid pretest problems.
See footnote 6.
To estimate the null model, we use the following matrix of instruments:
$$\begin{aligned} H_{0}=Q_{0}[X,X_{-1},P] \end{aligned}$$Results from the estimation of the null, alternative and augmented model are reported in the Appendix.
The weighting matrix employed in this paper is based on the six nearest neighbors.
There is an additional difference between the null and alternative model since the model under \(H_{1}\) is specified in terms of a spatially lagged dependent variable. This was brought to our attention by an anonymous referee. In light of this, we decided to run an additional Monte Carlo experiment where the competing models presented two differences: both choice of W and inclusion or exclusion of the spatially lagged dependent variable. To make this additional Monte Carlo experiment more similar to our empirical application, we set \(N=49\) and \(T=25\). The details on this experiment are not reported in the paper but are available from the authors. In summary, while the power is very high, for all parameters combination, estimates for the size are, on average, 0.038 and 0.039 for the two predictors.
Note that the spatial lag in the price variable is one of the columns in the spatial lag of M.
In the paper, we only present the test based on the predictor corresponding to the minimum information set. The results for the other predictor are qualitatively similar and, therefore, are only available upon request from the authors.
In the empirical application as well as in all of the Monte Carlo experiments in Sect. 8, the denominator of \(\tilde{\Omega }_{H^{\prime }Q_{0}RR^{\prime }Q_{0}H}\) is \(N(T-1)-k\) where k is the number of regressors in the model. This degrees of freedom correction is typically done in fixed effects studies (see, e.g., Baltagi 2008, equation 2.24). Also, in order to estimate the residuals more efficiently, \(Q_{0}u\) is estimated from the null model.
We generate the spatial panel data with \(100+T\) periods and then take the last T as our sample, and we set T equal to 5 in all experiments. The initial values are generated as
$$\begin{aligned} y_{0}=(I_{T}\otimes (I_{N}-\lambda _{0}W_{0})^{-1})X_{0}\beta _{0} \end{aligned}$$In a sense, this result is not very surprising given that the power of any test will depend on the extent to which the null and alternative hypotheses differ. A further discussion of this is given by LeSage and Pace (2009). Using Bayesian posterior model comparisons, they illustrate for alternative weight matrices that as the spatial dependence approaches low levels, the posterior probabilities approach the prior probabilities. In the limiting case, if the spatial dependence were zero, an empirical test would not be able to distinguish between two different weighting matrices.
However, as pointed out by ** and Lee (2013), there might be situations where the gap in the power for the two versions of the spatial J test may be large. This is an issue that could be explored in a larger Monte Carlo study, and we leave it for further research.
Some studies have suggested to implement bootstrap testing procedure to improve the small sample performance of the test (see, e.g., Burridge and Fingleton 2010, for an example in a spatial context). We decided to leave this for future research.
References
Anselin L (1986) Non-nested tests on the weight structure in spatial autoregressive models: some Monte Carlo results. J Reg Sci 26:267–284
Anselin L, Le Gallo J, Jayet H (2008) Spatial panel econometrics. In: Matyas L, Sevestre P (eds) The econometrics of panel data, fundamentals and recent developments in theory and practice, 3rd edn. Springer, Berlin, pp 624–660
Anselin L, Lozano-Gracia N (2008) Errors in variables and spatial effects in hedonic house price models of ambient air quality. Empirical Economics 34(1):5–34
Baltagi BH (2008) Econometric analysis of panel data, 4th edn. Wiley, New York
Baltagi BH, Egger P, Pfaffermayr M (2007a) Estimating models of complex FDI: are there third-country effects? J Econom 140:260–281
Baltagi BH, Egger P, Pfaffermayr M (2013) A generalized spatial panel data model with random effects. Econom Rev 32(5–6):650–685
Baltagi BH, Kelejian HH, Prucha IR (2007b) Analysis of spatially dependent data. J Econom 140:1–4
Baltagi BH, Levin D (1986) Estimating dynamic demand for cigarettes using panel data: the effects of bootlegging, taxation and advertising reconsidered. Rev Econ Stat 68:148–155
Baltagi BH, Levin D (1992) Cigarette taxation: raising revenues and reducing consumption. Struct Change Econ Dyn 3:321–335
Baltagi BH, Liu L (2008) Testing for random effects and spatial lag dependence in panel data models. Stat Probab Lett 78:3304–3306
Baltagi BH, Song SH, Jung BC, Koh W (2007c) Testing for serial correlation, spatial autocorrelation and random effects using panel data. J Econom 140(1):5–51
Baltagi BH, Song SH, Koh W (2003) Testing panel data regression models with spatial error correlation. J Econom 117:123–150
Burridge P, Fingleton B (2010) Bootstrap inference in spatial econometrics: the \(J\) test. Spat Econ Anal 5:93–119
Dastoor NK (1983) Some aspects of testing non-nested hypothesis. J Econom 21:213–228
Davidson R, MacKinnon J (1981) Several tests for model specification in the presence of alternative hypotheses. Econometrica 49:781–794
Debarsy N, Ertur C (2010) Testing for spatial autocorrelation in a fixed effects panel data model. Reg Sci Urban Econ 40:453–470
Debarsy N, Ertur C, LeSage JP (2012) Interpreting dynamic space-time panel data models. Stat Methodol 9:158–171
Delgado MA, Stengos T (1994) Semiparametric specification testing of non-nested econometric models. Rev Econ Stud 61:291–303
Elhorst JP (2001) Dynamic models in space and time. Geogr Anal 33(2):119–140
Elhorst JP (2003) Specification and estimation of spatial panel data models. Int Reg Sci Rev 26(3):244–268
Elhorst JP (2005) Unconditional maximum likelihood estimation of linear and log-linear dynamic models for spatial panels. Geogr Anal 37:62–83
Elhorst JP (2008) Serial and spatial error correlation. Econ Lett 100:422–424
Elhorst JP (2009) Spatial panel data models. In: Fischer MM, Getis A (eds) Handbook of applied spatial analysis. Springer, Berlin
Elhorst JP (2010) Dynamic panels with endogenous interactions effects when T is small. Reg Sci Urban Econ 40:272–282
Elhorst JP, Freret S (2009) Yardstick competition among local governments: French evidence using a two-regimes spatial panel data model. J Reg Sci 49:931–951
Elhorst JP, Piras G, Arbia G (2010) Growth and convergence in a multi-regional model with space-time dynamics. Geogr Anal 42:338–355
Florax RJGM, Folmer H, Rey SJ (2003) Specification searches in spatial econometrics: the relevance of Hendry’s methodology. Reg Sci Urban Econ 33:557–579
Godfrey L (1983) Testing non-nested models after estimation by instrumental variables or least squares. Econometrica 51:355–366
Godfrey L, Pesaran MH (1983) Test of non-nested regression models. J Econom 21:133–154
Greene W (2003) Econometric analysis, 5th edn. Prentice Hall, Upper Saddle River
** F, Lee LF (2013) Cox-type tests for competing spatial autoregressive models with spatial autoregressive disturbances. Reg Sci Urban Econ 43:590–616
Kapoor M, Kelejian HH, Prucha IR (2007) Panel data model with spatially correlated error components. J Econom 140(1):97–130
Kelejian HH (2015) Critical issues in spatial models: error term specifications, additional endogenous variables, pre-testing, and bayesian analysis. Lett Spat Resour Sci. doi:10.1007/s12076-015-0146-2
Kelejian HH (2008) A spatial \(J\)-test for model specification against a single or a set of nonnested alternatives. Lett Spat Resour Sci 1(1):3–11
Kelejian HH, Murrell P, Shepotylo O (2013) Spatial spillovers in the development of institutions. J Dev Econ 101:297–315
Kelejian HH, Piras G (2011) An extension of Kelejian’s J-test for non-nested spatial models. Reg Sci Urban Econ 41:281–292
Kelejian HH, Piras G (2015) An extension of the \(J\)-test to a spatial panel data framework. J Appl Econom. doi:10.1002/jae.2425
Kelejian HH, Prucha IR (1999) A generalized moments estimator for the autoregressive parameter in a spatial model. Int Econ Rev 40(2):509–533
Kelejian HH, Prucha IR (2007) HAC estimation in a spatial framework. J Econom 140:131–154
Kim MS, Sun Y (2011) Spatial heteroskedasticity and autocorrelation consistent estimation of covariance matrix. J Econom 160(2):349–371
Kmenta J (1986) Elements of econometrics, 2nd edn. Macmillan, New York
Lee LF, Yu J (2010a) Estimation of spatial autoregressive panel data models with fixed effects. J Econom 154:165–185
Lee LF, Yu J (2010b) Some recent development in spatial panel data models. Reg Sci Urban Econ 40:255–271
Lee LF, Yu J (2010c) A spatial dynamic panel data model with both time and individual fixed effects. Econom Theory 26:564–597
Lee LF, Yu J (2010d) A unified transformation approach to the estimation of spatial dynamic panel data models: stability, spatial cointegration and explosive roots. In: Ullah A, Giles D (eds) Handbook of empirical economics and finance. Chapman and Hall - CRC, Boca Raton, pp 397–434
LeSage JP, Pace K (2009) Introduction to spatial econometrics. CRC Press, Boca Raton, FL
MacKinnon J, White H, Davidson R (1983) Test for model specification in the presence of alternative hypotheses: some further results. J Econom 21(1):53–70
Mutl J, Pfaffermayr M (2011) The Hausman test in a Cliff and Ord panel model. Econom J 14:48–76
Parent O, LeSage JP (2010) A spatial dynamic panel model with random effects applied to commuting times. Transp Res Part B 44:633–645
Parent O, LeSage JP (2011) A space-time filter for panel data models containing random effects. Comput Stat Data Anal 55(1):475–490
Parent O, LeSage JP (2012) Spatial dynamic panel data models with random effects. Reg Sci Urban Econ 42(4):727–738
Pesaran HM, Tosetti E (2011) Large panels with common factors and spatial correlations. J Econom 161(2):182–202
Pesaran MH (1974) On the general problem of model selection. Rev Econ Stud 41:153–171
Pesaran MH (1982) Comparison of local power of alternative tests of non-nested regression models. Econometrica 50:1287–1305
Pesaran MH, Deaton AS (1978) Testing non-nested non-linear regression models. Econometrica 46:677–694
Pesaran MH, Weeks M (2001) Non-nested hypothesis testing: an overview. In: Baltagi BH (ed) A companion to theoretical econometrics. Blackwell, Malden
Piras G (2013) Efficient GM estimation of a cliff and ord panel data model with random effects. Spat Econ Anal 8(3):370–388
Pötscher BM, Prucha IR (2000) Basic elements of asymptotic theory. In: Baltagi BH (ed) A companion to theoretical econometrics. Basil Blackwell, Oxford, pp 201–229
White H (1980) A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48(4):817–838
Yu J, de Jong R, Lee LF (2008) Quasi maximum likelihood estimators for spatial dynamic panel data with fixed effects when both n and t are large. J Econom 146:118–134
Yu J, Lee LF (2010) Estimation of unit root spatial dynamic panel data models. Econom Theory 26:1332–1362
Author information
Authors and Affiliations
Corresponding author
Additional information
We would like to thank the referees and the editor Badi Baltagi for helpful comments on an earlier version of this paper. They are not, however, responsible for any shortcoming remaining in this paper.
Appendix
Appendix
1.1 Proof of Theorem 1 (Table 4)
To simplify notation, we prove Theorem 1 for the case in which \(\overbrace{ Q_{0}Y_{J}^{E,A}}\)is used, \(J=1,\ldots ,G.\) The proof for the case in which \( \overbrace{Q_{0}Y_{J}^{E,B}}\) is used is virtually identical.
First note from (10), (14) and (15) that
where \(Z_{1,G}=[Z_{1},\ldots ,Z_{G}]\) and \(\hat{\gamma }_{1,G}=[\hat{\gamma } _{1}^{\prime },\ldots ,\hat{\gamma }_{G}^{\prime }]^{\prime }\) where \(\hat{\gamma } _{J}^{\prime }=(\hat{\phi }_{1,J}^{\prime },\hat{\phi }_{2,J}^{\prime },\hat{ \lambda }_{J},\hat{\alpha }_{J}),\) \(J=1,\ldots ,G.\) For this case \(\hat{\Gamma } _{A} \) in (16) is
and \(\Gamma _{A}=[Z,Z_{1,G}\) \(\gamma _{1,G}]\). Therefore, recalling that \(\tilde{\Phi }_{A}=P_{H}\hat{\Phi }_{A}\) and \(\hat{\Phi }_{A}=Q_{0}\hat{\Gamma } _{A}\) it follows from (16) and (18) that
Consider the term \((NT)^{-1}H^{\prime }Q_{0}\hat{\Gamma }_{A}\) in the inverse on the first line of (31) and note that by Assumption 5, \(\hat{\gamma }_{1,G}\overset{P}{\rightarrow }C_{1,G}=(c_{1}^{\prime },\ldots ,c_{G}^{\prime })^{\prime },\) where \(c_{J}^{\prime }=\) \( (c_{0,J},c_{2,J},L_{J},a_{J}),J=1,\ldots ,G\). Also note that \(Q_{0}^{\prime }=Q_{0},Q_{0}^{2}=Q_{0},\) and therefore from (17) \(Q_{0}H=H\) and so from Part (b) of Assumption 6, and (29)
Let \(\hat{\digamma }\) be the inverse term on the first line of (31) and let
Then from (32), and parts (a) and (b) of Assumption 6, first note that
where \(\Omega _{H\Gamma _{A}}^{\prime }\Omega _{HH}^{-1}\Omega _{H\Gamma _{A}}\) is positive definite and, therefore, non-singular since \(\Omega _{HH}\) is positive definite and so, therefore, is \(\Omega _{HH}^{-1},\) and \(\Omega _{H\Gamma _{A}}\) has full column rank. It then follows from (32) to (34), and parts (a) and (b) of Assumption 6 that
Finally, consider the last term on the second line in (31), namely \( (NT)^{-1/2}H^{\prime }Q_{0}R\varepsilon .\) Since the row and column sums of both \(Q_{0}\) and R are uniformly bounded in absolute value, the row and column sums of \(Q_{0}R\) are also uniformly bounded in absolute value. Since by Assumption 4 (a) the elements of H are uniformly bounded in absolute value, it follows that the elements of \(H^{\prime }Q_{0}R=H^{\prime }R\) are uniformly bounded in absolute value. Given this, and Assumptions 1, 6 part (c) and the central limit theorem (30) in Pötscher and Prucha (2000), it follows that
Therefore, by the continuous map** theorem and (31)–(35)
Rights and permissions
About this article
Cite this article
Kelejian, H.H., Piras, G. A J test for dynamic panel model with fixed effects, and nonparametric spatial and time dependence. Empir Econ 51, 1581–1605 (2016). https://doi.org/10.1007/s00181-015-1056-2
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00181-015-1056-2