Skip to main content

Measuring productivity and efficiency: a Kalman filter approach


In the Kalman filter setting, one can model the inefficiency term of the standard stochastic frontier composed error as an unobserved state. In this study a panel data version of the local level model is used for estimating time-varying efficiencies of firms. We apply the Kalman filter to estimate average efficiencies of U.S. airlines and find that the technical efficiency of these carriers did not improve during the period 1999–2009. During this period the industry incurred substantial losses, and the efficiency gains from reorganized networks, code-sharing arrangements, and other best business practices apparently had already been realized.

This is a preview of subscription content, access via your institution.

Fig. 1


  1. See also Galán and Pollitt (2014) for an empirical study.

  2. See, also Sickles (2005).

  3. Our paper is a substantially revised and extended version of Chapter 2 in Levent Kutlu’s dissertation, Market Power and Efficiency (2010a). Recently, independent from us, Peyrache and Rambaldi (2013) proposed a similar Kalman filter model for estimating efficiencies.

  4. For details see Jin and Jorgenson (2009), Kim (2006), Kim and Kim (2011), Kim and Nelson (2006), and Kutlu and Sickles (2012).

  5. In the Kalman filter setting it is possible to estimate a cost function with/without input share equations. For the simultaneous equations setting we do not consider a stochastic frontier model because of so called Greene’s problem. See Kumbhakar (1997), Kumbhakar and Lovell (2003), and Kutlu (2013).

  6. See Durbin and Koopman (2001) for more details about initialization.

  7. See Durbin and Koopman (2001) and Kutlu and Sickles (2012) for details of the square-root Kalman filter.

  8. See Appendix 1 for more details about required degrees of freedom. The degrees of freedom requirement may be eased by using other (yet restrictive) initialization approaches.

  9. We failed to estimate the full model for this short panel data.

  10. Kutlu (2010b), Karakaplan and Kutlu (2015), and Tran and Tsionas (2012) use similar control function approaches to deal with endogeneity issues in the stochastic frontier context.

  11. See Harvey (1989) for more details on this type of solutions to the endogeneity problem in the Kalman filter setting.

  12. When generating regressors we followed Park et al. (2003, 2007) and Kutlu (2010b).

  13. Note that for the KFE μ it may be negative or positive. Hence, as long as the u it term is predicted properly the sign of μ it is not important. However, in the simulations the production model is written in a general form so that BC production model is also nested. Hence, for this purpose the sign of μ it is negative.

  14. The original CSSW estimator assumes W it  = [1, t, t 2]. However, for the simulations we normalize t by n t . This normalization does not affect the results and is done for numerical purposes.

  15. In some of the simulation runs we observed even negative correlations.

  16. For more information about the financial situations of U.S. airlines see Borenstein (2011).

  17. The full data set has 39 airlines and 321 observations. We droped 1 airline with less than 4 observations and 3 cargo airlines.

  18. As mentioned in our theoretical section we want to concentrate on the simple local level model rather than the full model as it is easier to estimate. For example, our attempt to estimate the full model failed. The alternative estimates for random walk with deterministic trend and AR(1) effects models are provided in Appendix 2.

  19. When calculating the efficiency estimates, we trim the effects term from the upper and lower 7.5 % percentiles, observed at least at one time period, to remove the outlier effects. See, Berger (1993), Berger and Hannan (1998), Kutlu (2012), and Kneip et al. (2012) for more details. See, also Appendix 2 for some robustness check for trimming.

  20. For similar results see also Diewert and Fox (2008).

  21. Note that smoothing is not needed to get the MLE estimates. The smoothing equations are calculated after the estimations. The Kalman filter uses past and current observations to predict the state variables; and thus it does not use all information when calculating the state variable predictions. Once the parameters of the model are estimated by MLE, the smoothing enables us to update our predictions using information from all time periods. This is why after smoothing procedure the predictions of state variables look smoother. Hence, the name smoothing.

  22. The initial values for the AR(1) model are estimated as parameters.


  • Ahn SC, Good DH, Sickles RC (2000) Estimation of long-run inefficiency levels: a dynamic frontier approach. Econ Rev 19:461–492

    Article  Google Scholar 

  • Ahn SC, Lee YH, Schmidt P (2013) Panel data models with multiple time-varying individual effects. J Econ 174:1–14

    Article  Google Scholar 

  • Aigner DJ, Lovell CAK, Schmidt P (1977) Formulation and estimation of stochastic frontier production functions. J Econ 6:21–37

    Article  Google Scholar 

  • Assaf AG, Gillen D, Tsionas EG (2014) Understanding relative efficiency among airports: a general dynamic model for distinguishing technical and allocative efficiency. Transp Res Part B 70:18–34

    Article  Google Scholar 

  • Battese GE, Coelli TJ (1992) Frontier production functions, technical efficiency and panel data with application to paddy farmers in India. J Prod Anal 3:153–169

    Article  Google Scholar 

  • Basu S, Fernald JG (1997) Returns to scale in U.S. production: estimates and implications. J Polit Econ 105:249–83

    Article  Google Scholar 

  • Berger AN (1993) Distribution free estimates of efficiency in US banking industry and tests of the standard distributional assumptions. J Prod Anal 4:261–292

    Article  Google Scholar 

  • Berger AN, Hannan TH (1998) The efficiency cost of market power in the banking industry: a test of the “Quiet Life” and related hypotheses. Rev Econ Stat 80:454–465

    Article  Google Scholar 

  • Borenstein S (2011) On the persistent financial losses of U.S. Airlines: a preliminary exploration, Working paper

  • Cornwell C, Schmidt P, Sickles RC (1990) Production frontiers with time-series variation in efficiency levels. J Econ 46:185–200

    Article  Google Scholar 

  • Desli E, Ray SC, Kumbhakar SC (2003) A dynamic stochastic frontier production model with time-varying efficiency. Appl Econ Lett 10:623–626

    Article  Google Scholar 

  • Diewert WE, Fox KJ (2008) On the estimation of returns to scale, technical progress and monopolistic markups. J Econ 145:174–193

    Article  Google Scholar 

  • Durbin J, Koopman SJ (2001) Time series analysis by state space methods, oxford statistical series 24. Oxford University Press, Oxford

    Google Scholar 

  • Emvalomatis G, Stefanou SE, Lansink AO (2011) A reduced-form model for dynamic efficiency measurement: application to dairy farms in Germany and The Netherlands. Am J Agric Econ 93:161–174

    Article  Google Scholar 

  • Färe R, Grosskopf S, Sickles RC (2007) Productivity? of U.S. Airlines after deregulation. J Transp Econ Policy 41:93–112

    Google Scholar 

  • Galán JE, Pollitt MG (2014) Inefficiency persistence and heterogeneity in Colombian electricity utilities. Energy Econ 46:31–44

    Article  Google Scholar 

  • Harvey AC (1989) Forecasting, structural time series models and the Kalman filter. Cambridge University Press, Cambridge

    Google Scholar 

  • Huang TH, Chen YH (2009) A study on long-run inefficiency levels of a panel dynamic cost frontier under the framework of forward-looking rational expectations. J Bank Financ 33:842–849

    Article  Google Scholar 

  • Jin H, Jorgenson DW (2009) Econometric modeling of technical change. J Econ 157:205–219

    Article  Google Scholar 

  • Jondrow J, Lovell CAK, Materov IS, Schmidt P (1982) On the estimation of technical inefficiency in the stochastic frontier production function model. J Econ 19:233–238

    Article  Google Scholar 

  • Kalman RE (1960) A new approach to linear filtering and prediction problems. J Basic Eng Trans ASMA Ser D 82:35–45

    Article  Google Scholar 

  • Karakaplan MU, Kutlu L (2015) Consolidation policies and saving reversals. Working paper (SSRN:

  • Kim CJ (2006) Time-varying parameter models with endogenous regressors. Econ Lett 91:21–26

    Article  Google Scholar 

  • Kim CJ (2008) Dealing with endogeneity in regression models with dynamic coefficients. Found Trends Econ 3:165–266

    Article  Google Scholar 

  • Kim CJ, Nelson CR (2006) Estimation of a forward-looking monetary policy rule: a time-varying parameter model using ex post data. J Monet Econ 53:1949–1966

    Article  Google Scholar 

  • Kim Y, Kim CJ (2011) Dealing with endogeneity in a time-varying parameter model: joint estimation and two-step estimation procedures. Econ J 14:487–497

    Google Scholar 

  • Kneip A, Sickles RC, Song W (2012) A new panel data treatment for heterogeneity in time trends. Econ Theory 28:560–628

    Article  Google Scholar 

  • Kumbhakar SC (1990) Production frontiers, panel data, and time-varying technical inefficiency. J Econ 46:201–211

    Article  Google Scholar 

  • Kumbhakar SC (1997) Modeling allocative inefficiency in a translog cost function and cost share equations: an exact relationship. J Econ 76:351–356

    Article  Google Scholar 

  • Kumbhakar SC, Lovell CAK (2003) Stochastic frontier analysis. Cambridge University Press, Cambridge

    Google Scholar 

  • Kutlu L (2010a) Battese–Coelli estimator with endogenous regressors. Econ Lett 109:79–81

    Article  Google Scholar 

  • Kutlu L (2010b) Market power and efficiency. Unpublished PhD dissertation, Rice University

  • Kutlu L (2012) U.S. banking efficiency, 1984–1995. Econ Lett 117:53–56

    Article  Google Scholar 

  • Kutlu L (2013) Misspecification in allocative inefficiency: a simulation study. Econ Lett 118:151–154

    Article  Google Scholar 

  • Kutlu L, Sickles RC (2012) Estimation of market power in the presence of firm level inefficiencies. J Econ 168:141–155

    Article  Google Scholar 

  • Lee YH, Schmidt P (1993) A production frontier model with flexible temporal variation in technical efficiency. Meas Prod Effic Tech Appl 237–255

  • Meeusen W, van den Broeck J (1977) Efficiency estimation from Cobb–Douglas production functions with composed error. Int Econ Rev 2:435–444

    Article  Google Scholar 

  • Park BU, Sickles RC, Simar L (2003) Semiparametric efficient estimation of AR1 panel data models. J Econ 117:279–309

    Article  Google Scholar 

  • Park BU, Sickles RC, Simar L (2007) Semiparametric efficient estimation of dynamic panel data models. J Econ 136:281–301

    Article  Google Scholar 

  • Peyrache A, Rambaldi AN (2013) Incorporating temporal and country heterogeneity in growth accounting—an application to EU-KLEMS, Working paper

  • Pitt MM, Lee LF (1981) The measurement and sources of technical inefficiency in the indonesian weaving industry. J Dev Econ 9:43–64

    Article  Google Scholar 

  • Schmidt P, Sickles RC (1984) Production frontiers and panel data. J Business Econ Stat 2:367–374

    Google Scholar 

  • Sickles RC (1985) A nonlinear multivariate error-components analysis of technology and specific factor productivity growth with an application to the U.S. Airlines. J Econ 27:61–78

    Article  Google Scholar 

  • Sickles RC (2005) Panel estimators and the identification of firm-specific efficiency levels in parametric, semiparametric, and nonparametric settings. J Econ 126:305–334

    Article  Google Scholar 

  • Sickles RC, Good DH, Johnson RL (1986) Allocative distortions and the regulatory transition of the U.S. Airline industry. J Econ 33:143–163

    Article  Google Scholar 

  • Sickles RC, Hao J, Sheng C (2015) Panel data and productivity measurement. In: Baltagi B (ed) Oxford handbook of panel data Chapter 17. Oxford University Press, New York, pp 517–547

  • Slade ME (1989) Modeling stochastic and cyclical components of technical change: an application of the Kalman filter. J Econ 41:363–383

    Article  Google Scholar 

  • Tran KC, Tsionas EG (2012) GMM estimation of stochastic frontier model with endogenous regressors. Econ Lett 118:233–236

    Article  Google Scholar 

  • Tsionas EG (2006) Inference in dynamic stochastic frontier models. J Appl Econ 21:669–676

    Article  Google Scholar 

  • Ueda T, Hoshino K (2005) Estimation of firms efficiencies using kalman filter and stochastic efficiency model. J Oper Res Soc Jpn. 308–317

Download references

Author information

Authors and Affiliations


Corresponding author

Correspondence to Levent Kutlu.

Ethics declarations

Conflict of interest

The authors declare that they have no conflict of interests.


Appendix 1

In this appendix we provide further details about the Kalman filter estimation. Consider two stochastic frontier models that are nested by the general setting that we described:

$$\begin{array}{rcl}{y_{it}} &=& {X_{it}}\beta + Z{B_{it}} + {\varepsilon _{it}},\quad {\varepsilon _{it}} \sim {\bf NID}( 0,{\sigma _\varepsilon ^2} )\\ {B_{i,t + 1}} &=& T{B_{i,t}} + R{e_{it}},\quad {e_{it}} \sim {\bf NID}( 0,{\sigma _e^2} ) \end{array}$$


$${B_{it}} = \displaystyle\left[\begin{array}{c}{\mu _{it}} \\ {\tau _{it}}\end{array}\right],\ T = \displaystyle\left[\begin{array}{cc} 1 & 1 \\ 0 & 1\end{array} \right],\,{\rm and}\ Z = \Big[ 1 \quad 0 \Big] $$


$$\begin{array}{rcl} {y_{it}} &=& {X_{it}}\beta + Z{B_{it}} + {\varepsilon _{it}},\quad {\varepsilon _{it}} \sim {\bf{NID}}\left( {0,\sigma _\varepsilon ^2} \right)\\ {B_{i,t + 1}} &=& T{B_{i,t}} + {e_{it}},\quad {e_{it}} \sim {\bf{NID}}\left( {0,\sigma _e^2} \right)\end{array}$$


$${B_{it}} = {\mu _{it}},\,T = \rho ,\,\,{\rm{and}}\,\,Z = 1$$

The first model assumes a random walk with deterministic trend effects term and the second model assumes a potentially non-stationary AR(1) process for the effects term. Both of these models can be estimated by using the recursive equations provided below. The estimation consists of two steps. Kalman filtering and smoothing. In the first step, the following recursive Kalman filter equations are applied:

$$\begin{array}{rcl} {\eta _{it}} &=& {y_{it}} - {X_{it}}\beta - Z{b_{it}}\\ \ \ \ \, {F_{it}} &=& Z{P_{it|t - 1}}Z\prime + \sigma _\varepsilon ^2\\ \ \ \ {M_{it}} &=& {P_{it|t - 1}}Z\prime \\ \kern0.8pc {b_{it|t}} &=& {b_{it|t - 1}} + {M_{it}}F_{it}^{ - 1}{\eta _{it}}\\ \ \ \, {P_{it|t}} &=& {P_{it|t - 1}} - {M_{it}}F_{it}^{ - 1}{{M'}_{it}}\\ {b_{it|t - 1}} &=& T{b_{i,t - 1|t - 1}}\\ {P_{it|t - 1}} &=& T{P_{i,t - 1|t - 1}}T\prime + \sigma _e^2 \end{array}$$

In the second step, the smoothing is applied by using the following recursive equations:

$$\begin{array}{rcl} \kern1pc{L_{it}} &=& T - T{M_{it}}F_{it}^{ - 1}Z\\ \kern0.2pc{r_{i,t - 1}} &=& Z\prime F_{it}^{ - 1}{\eta _{it}} + {{L'}_{it}}{r_{it}}\\ {N_{i,t - 1}} &=& Z\prime F_{it}^{ - 1}Z + {{L'}_{it}}{N_{it}}{L_{it}}\\ {{\tilde b}_{it|t - 1}} &=& {b_{it|t - 1}} + {P_{it|t - 1}}{r_{i,t - 1}}\\ \kern0.9pc{V_{it}} &=& {P_{it|t - 1}} - {P_{it|t - 1}}{N_{i,t - 1}}{P_{it|t - 1}}\end{array}$$

where \({r_{i{n_t}}}\, = \,0\) and \({N_{i{n_t}}}\, = \,0\).Footnote 21 The log-likelihood is given by:

$${ln\,(L) = {\sum}_{i\, = \,1}^{n_i} {L_i} = {\rm constant} - \frac{1}{2}\mathop {\sum}\limits_{i\, = \,1}^{n_i} \mathop {\sum}\limits_{t\, = \,{d_i}\, + \,1}^{n_t} \left( {ln\left( {F_{it}} \right) + \frac{{\eta _{it}^2}}{{F_{it}}}} \right)}$$

where d i is the number of diffuse states for firm i. The number of diffuse priors (per panel unit) for the first model is two. The number of diffuse priors (per panel unit) for the second model is one. If we assume that ρ = 1, the second model would still have one diffuse prior per panel unit. However, obviously, the number of parameters to be estimated would be smaller. If we assume that |ρ| < 1, the second model would not have any diffuse priors.

Let m be the number of state variables (per panel unit) and q be the number of state variables with diffuse priors (per panel unit). For diffuse initialization we assume that:

$${b_{i0}} = A\delta + S{\eta _{i0}}$$
$${\eta _{i0}} \sim {\bf{NID}}\left( {0,{Q_0}} \right)$$

where δ is a q × 1 vector of unknown quantities and the m × q matrix A and m × (mq) matrix S are selection matrices that consist of columns of identity matrix. Then, matrix for initialization is:

$${P_{i0}} = \kappa {P_{i\infty 0}} + {P_{ * 0}}$$

where κ→∞, P ∞ = AA, and P * = SQ 0 S′. As it can be seen from the log-likelihood, the first d i observation(s) for panel unit i are burnt out for the sake of initialization and are not considered in the log-likelihood. Hence, for example, for the second model the first observation of each panel unit is used for initialization. The reason for this is that as long as t ≥ d i  + 1, we would have P it  = 0. The variance matrix can be estimated using the standard maximum likelihood procedures. For the estimations we used the standard BFGS optimization method.

Appendix 2

In this appendix we present additional results based on the full translog model and our truncation scheme when calculating the efficiency estimates for KFE and CSSW estimator. We also provide estimates for alternative Kalman filter models.

The full translog estimates are given in Table 10. The parameter estimates are generally not significant even at 10 % significance level. The median of the returns to scale values for the KFE, CSSW, and BC estimators are 0.8625, 1.1478, and 1.0184, respectively. The corresponding returns to scale estimates from the restricted model were 0.883, 0.94, and 1.034, respectively. Hence, for the KFE and BC estimator the returns to scale estimates are robust to the choice of the functional form. Nevertheless, for both restricted and unrestricted translog production models the constant returns to scale value of 1 lies within one sample standard deviation away from the median value of returns to scale estimates from each of these estimates. All the estimators satisfy the monotonicity conditions at the median values of the regressors at each time period. In contrast to the restricted translog production model where only KFE satisfied the regularity conditions at the median values of the regressors, KFE and CSSW estimator satisfies the curvature conditions at each time period. BC estimator failed to satisfy the regularity conditions at four of the time periods. The estimates for the production function parameters and average efficiencies for the KFE and the BC and CSSW estimators are given in Table 10 and Fig. 2. The overall average efficiencies for the KFE, CSSW, and BC are 0.637, 0.458, and 0.605, respectively. These values are not substantially different from their restricted counterparts, i.e., 0.577, 0.438, and 0.632. The average efficiencies for the full translog model are provided in Fig. 2. In line with the restricted translog model, KFE predicts decrease in efficiency in first few years of the study period and relatively stable efficiency levels for the last couple of years.

Table 10 Full translog production model estimates
Fig. 2
figure 2

Efficiency estimates for translog model

Now, we present the efficiency estimates when the trimming for KFE and CSSW are done for top–bottom 5 % (rather than 7.5 %) of the effects term when calculating the efficiencies. The BC estimates remain the same as they are not subject to such trimming. The average efficiency estimates for 5 % trimming case are provided in Figs. 3 and 4.

Fig. 3
figure 3

Efficiency estimates for restricted translog model

Fig. 4
figure 4

Efficiency estimates for translog model

Figure 5 presents the average of efficiency estimates for restricted translog model. Due to outliers the KFE and CSS model estimates are low.

Fig. 5
figure 5

Efficiency estimates for restricted translog model without trimming

Finally, we provide our estimates for alternative Kalman filter models in Table 11. As we mentioned we failed to estimate the full Kalman filter model that we presented in Eq. 1. We rather estimated the models given in Eq. 12 (random walk model with deterministic trend) and Eq. 13 (AR(1) model).Footnote 22 Based on the BIC values the random walk model with deterministic trend is not preferred. In particular, BIC values for random walk with deterministic trend and random walk models are 2.1 vs. 1.7, respectively. The second model seems to be subject to the pile up problem as one of the variance parameters is collapsed to zero. Hence, we only provide the results for the sake of completeness. Nevertheless, the estimate of ρ = 1.0315 parameter indicates that our random walk assumption for the trend term is sensible for our empirical example. Hence, these findings support our choice for using the random walk model as our benchmark model.

Table 11 Alternative Kalman filter model estimates

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Duygun, M., Kutlu, L. & Sickles, R.C. Measuring productivity and efficiency: a Kalman filter approach. J Prod Anal 46, 155–167 (2016).

Download citation

  • Published:

  • Issue Date:

  • DOI:


  • Kalman filter
  • Panel data
  • Airline productivity

JEL Classification

  • C13
  • C23
  • R49