Skip to main content
SpringerLink
Log in

Density deconvolution with Laplace errors and unknown variance

  • Published:
Journal of Productivity Analysis Aims and scope Submit manuscript

Abstract

We consider density deconvolution with zero-mean Laplace noise in the context of an error component regression model. We adapt the minimax deconvolution methods of Meister (2006) to allow estimation of the unknown noise variance. We propose a semi-uniformly consistent estimator for an ordinary-smooth target density and a modified "variance truncation device” for the unknown noise variance. We provide a simulation study and practical guidance for the choice of smoothness parameters of the ordinary-smooth target density. We apply restricted versions of our estimator to a stochastic frontier model of US banks and to a measurement error model of daily saturated fat intake.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3
Fig. 4

Similar content being viewed by others

Notes

  1. See Butucea and Matias (2005), Matias (2002), Meister (2006), Schwartz and Bellegem (2010), and Horrace and Parmeter (2011). See Delaigle and Van Keilegom (2021) for a good survey.

  2. The Meister (2006) estimator is uniformly consistent relative to the target distributional family but individually relative to the noise distributional family. That is, consistency of the estimator does not hold uniformly over all noise distributions in the super-smooth family.

  3. Horrace and Parmeter (2018) perform maximum likelihood estimation of the parametric stochastic frontier model, not deconvolution.

  4. Neumann (1997), Johannes (2009), and Wang and Ye (2012) study deconvolution with fully unknown error distribution but require either an additional sample of the error or repeated observations, yjt.

  5. When u is not from a parametric family, analogous sufficient conditions are that the density of u is asymmetric and does not contain a “symmetric component” (see Delaigle and Hall 2016). Since the Laplace density is symmetric, our model is actually nonparametrically identified, if we further assume that the density of u is asymmetric, a common parametric assumption in the stochastic frontier literature.

  6. Schawrz and Van Bellegem (2010) also appeal to stability of the Normal distribution to identify the noise variance and a target density that vanishes on an interval of positive Lebesgue measure when noise is Normal. We note that their identification condition on the target density is fully compatible with the usual assumptions in the stochastic frontier model

  7. Stable distributional families are closed to self-convolution, while the logarithm of geometric stable distributional families are closed to self-convolution.

  8. Alternatively, b2 is identified if the density of u is asymmetric without a symmetric component, per Delaigle and Hall (2016), but here we prefer the target density be potentially symmetric.

  9. Indeed, the deconvolution estimator of Dattner and Goldenshluger (2011) relies on very general classes of distributions for the target and noise densities that includes the Laplace–Laplace convolution as a special case, and identification and consistent target density estimation are achieved for the known noise variance case.

  10. Butucea and Matias (2005) develop a generalized semiparametric deconvolution estimator, where the target density is ordinary smooth and the noise is super-smooth, a generalization of the parametric assumption of Normal noise. Without the parametric assumption, identification requires that the tails of the noise characteristic function decay faster than the those of the target density. We need no such assumption here, because we select a parametric family of distributions for noise.

  11. In the section “Some useful extensions”, we propose setting \({w}_{n}={k}_{n}/{\rm{ln}}{k}_{n}\) in the case C1 and δ are not fully known.

  12. Recall that for a Laplace distribution as defined in Assumption 2, the variance is 2b2. Hence, a natural upper bound for b2 is one-quarter the variance.

  13. Similar to Meister (2006), it does not leverage the relative decay rates of the noise and target characteristic functions.

  14. In Meister (2006), the bounding of the Normal variance is what leads to semi-uniformly consistency (as opposed to uniform consistency). Here, for Laplace errors, we still impose this "strong” condition for ease of proof. However, it may not be a necessary condition given Assumption 3.

  15. The compound effect of estimating the regression function will slow the target density rate compared to pure (non-regression) deconvolution, but the final rate is not a simple algebraic sum of the rates.

  16. … and the estimator of Meister (2006) as well.

  17. Actually, if one wants to assume the random noise is super-smooth with similarity index s, the smoothness parameter δ of target density can be estimated as well as the s by an adaptive procedure proposed by Butucea et al. (2008).

  18. Details see Condition 2.1 in Li and Racine (2007).

  19. The data are publicly available on the Journal of Applied Econometrics data archive website http://qed.econ.queensu.ca/jae/2009-v24.1/feng-serletis/.

  20. Once \({\hat{f}}_{u}\) is obtained, one can estimate the efficiency score using numerical integration on a grid of \(\hat{\varepsilon }\). To avoid an overloading of present paper, we stick to the estimation of marginal density of u.

  21. It is interesting to note that with a skew of 1.55, this provides evidence against use of the Half-Normal distribution (Papadopoulos and Parmeter 2021).

  22. For the Laplace distribution, δ = 2; for convolved Laplace, δ = 4. The choice δ = 3 is between Laplace and convolved Laplace.

  23. For Laplace deconvolution, we can apply directly Example 1 in Zhang and Karunramuni (2000).

  24. We use the package "ksdensity” in Matlab for the ErrorFree case.

  25. We also tried larger range of δ and narrow down to this specific range by searching the minimum of Δ.

  26. It seems to violate the independence assumption between the target variable and the measurement error.

  27. Since the chosen smooth parameter δ = 1.5 which is different from a Laplace distribution, the proposed method is still valid even with Var(u) < Var(v) as is explained in the paragraph following Assumption 3.

  28. Under Assumption 1, i.e., u and v are independent, the variance of Y should be the sum of the variances of u and v. Empirically, this may not be the case for real data.

  29. One point worth mentioning is that these minimax deconvolution techniques can produce error variance estimates equal to zero as we vary the choice of C1 and δ. Recall that \({\hat{b}}_{n}^{2}\) is bound between 0 and \(0.25V(\hat{\varepsilon })\). When it happens, the deconvolution estimators will be very similar to the ErrorFree estimator.

References

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

  • Butucea C, Matias C (2005) Minimax estimation of the noise level and of the deconvolution density in a semiparametric convolution model. Bernoulli 309–340

  • Butucea C, Matias C, Pouet C (2008) Adaptivity in convolution models with partially known noise distribution. Electron J Stat 897–915

  • Cao CD (2016) Linear regression with Laplace measurement errors. Master thesis. Department of Statistic, Kansas State University

  • Carroll RJ, Ruppert D, Stefanski LA, Crainiceanu CM (2006) Measurement error in nonlinear models: a modern perspective, vol 2E. CRC Press, Boca Raton, FL

  • Dattner AI, Goldenshluger AJ (2011) On deconvolution of distribution functions. Ann Stat 39:2477–2501

    Article  Google Scholar 

  • Delaigle A, Gijbels I (2004) Practical bandwidth selection in deconvolution kernel density estimation. Comput Stat Data Anal 45(2):249–267

    Article  Google Scholar 

  • Delaigle A, Hall P (2016) Methodology for non-parametric deconvolution when the error distribution is unknown. J Royal Stat Soc, Ser B, 78:231–252

  • Delaigle A, Van Keilegom I (2021) Deconvolution with unknown error distribution. In: Yi G, Delaigle A, Gustafson P (eds) Handbook of “Measurement Error Models”. CRC press, Boca Raton, FL, 245–270. https://doi.org/10.1201/9781315101279-12

  • Evdokimov K (2010) Identification and estimation of a nonparametric panel data model with unobserved heterogeneity. Working paper version

  • Fan J (1991) On the optimal rates of convergence for nonparametric deconvolution problems. Ann Stat 19:1257–1272

  • Fan Y, Li Q, Weersink A (1996) Semiparametric estimation of stochastic production frontier models. J Bus Econ Stat 14:460–468

  • Feng GH, Serletis A (2009) Efficiency and productivity of the US banking industry, 1998-2005: evidence from the Fourier cost function satisfying global regularity conditions. J Appl Eco 24:105–138

    Article  Google Scholar 

  • Greene WH (1990) A Gamma-distributed stochastic frontier model. J Econom 46(1-2):141–164

    Article  Google Scholar 

  • Greene WH (2003) Simulated likelihood estimation of the Normal-Gamma stochastic frontier function. J Prod Anal 19(2):179–190

    Article  Google Scholar 

  • Hall P, Simar L (2002) Estimating a changepoint, boundary, or frontier in the presence of observation error. J Am Stat Assoc 97:523–534

  • Henderson DJ, Parmeter CF (2015) Applied nonparametric econometrics. Cambridge University Press

  • Horowitz JL, Markatou M (1996) Semiparametric estimation of regression models for panel data. Rev Econ Stud 63:145–168

    Article  Google Scholar 

  • Horrace WC (2015) Moments of the truncated normal distribution. J Prod Anal 43:133–138

    Article  Google Scholar 

  • Horrace WC, Parmeter CF (2011) Semiparametric deconvolution with unknown error variance. J Prod Anal 35:129–141

    Article  Google Scholar 

  • Horrace WC, Parmeter CF (2018) A Laplace stochastic frontier model. Econom Rev 37:260–280

    Article  Google Scholar 

  • Jirak M, Meister A, Reiss M (2014) Adaptive function estimation in nonparametric regression with one-sided errors. Ann Stat 42(5):1970–2002

    Article  Google Scholar 

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

    Article  Google Scholar 

  • Jones DY et al. (1987) Dietary fat and breast cancer in the National Health and Nutrition Examination Survey I Epidemiologic Follow-up Study. J Natl Cancer Inst 79(3):465–471

    Google Scholar 

  • Johannes J (2009) Deconvolution with unknown error distribution. Ann Stat 37:2301–2323

    Article  Google Scholar 

  • Ju G, Gan L, Li Q (2019) Nonparametric panel estimation of labor supply. J Bus Econ Stat 37(2):260–274

    Article  Google Scholar 

  • Koul HL, Song WX (2014) Simulation extrapolation estimation in parametric models with Laplace measurement error. Electro J Stat 8:1973–1995

    Google Scholar 

  • Kneip A, Simar L, Van Keilegom I (2015) Frontier estimation in the presence of measurement error with unknown variance. J Econom 184(2):379–393

    Article  Google Scholar 

  • Kumbhakar SC, Tsionas EG (2005) Measuring technical and allocative inefficiency in the translog cost system: a Bayesian approach. J Econom 126:355–384

    Article  Google Scholar 

  • Li Q, Racine JS (2007) Nonparametric econometrics: theory and practice. Princeton University Press, Princeton, NJ

  • Matias C (2002) Semiparametric deconvolution with unknown noise variance. ESAIM: Probab Stat 6:271–292

  • Meeusen W, van den Broeck J (1977) Efficiency estimation from Cobb-Douglas production functions with composed error. Int Econ Rev 18(2):435–444

    Article  Google Scholar 

  • Meister A (2004) On the effect of misspecifying the error density in a nonparametric deconvolution problem. Can J Stat 32:439–449

    Article  Google Scholar 

  • Meister A (2006) Density estimation with normal measurement error with unknown variance. Stat Sin 16:195–211

    Google Scholar 

  • Meister A (2009) Deconvolution problems in nonparametric statistics. Lecture Notes in Statistics, Springer, Berlin. p, 193

  • Neumann MH (1997) On the effect of estimating the error density in nonparametric deconvolution. J Nonparametr Stat 7:307–330

    Article  Google Scholar 

  • Nguyen NB (2010) Estimation of technical efficiency in stochastic frontier analysis. PhD thesis, Bowling Green State University

  • Papadopoulos A, Parmeter CF (2021) Type II failure and specification testing in the stochastic frontier model. Eur J Oper Res 293(3):990–1001

    Article  Google Scholar 

  • Parmeter CF, Wang HJ, Kumbhakar SC (2017) Nonparametric estimation of the determinants of inefficiency. J Prod Anal 47:205–221

    Article  Google Scholar 

  • Parzen E (1962) On estimation of a probability density function and mode. Ann Math Stat 35:1065–1076

    Article  Google Scholar 

  • Ritter C, Simar L (1997) Pitfalls of Normal-Gamma stochastic frontier models. J Prod Anal 8(2):167–182

    Article  Google Scholar 

  • Schwarz M, Van Bellegem S (2010) Consistent density deconvolution under partially known error distribution. Stat Probab Lett 80:236–241

  • Simar L, van Keilegom I, Zelenyuk V (2017) Nonparametric least squares methods for stochastic frontier models. J Prod Anal 47:189–204

    Article  Google Scholar 

  • Song WX, Shi JH, Zhang CX (2016) Linear errors-in-variables regression and Tweedie’s formula. Working paper

  • Stefanski LA, Carroll RJ (1990) Deconvoluting kernel density estimators Statistics 21:169–184

    Article  Google Scholar 

  • Stevenson R (1980) Likelihood functions for generalized stochastic frontier estimation. J Econom 13(1):58–66

    Article  Google Scholar 

  • Tsionas MG (2017) “When, where, and how” of efficiency estimation: Improved procedures for stochastic frontier modeling. J Amer Stat Assoc 112:948–965

  • Wang WS, Amsler C, Schmidt P (2011) Goodness of fit tests in stochastic frontier models. J Prod Anal 35:95–118

    Article  Google Scholar 

  • Wang WS, Schmidt P (2009) On the distribution of estimated technical efficiency in stochastic frontier models. J Econom 148:36–45

    Article  Google Scholar 

  • Wang XF, Ye D (2012) The effects of error magnitude and bandwidth selection for deconvolution with unknown error distribution. J Nonparametr Stat 24:153–167

    Article  Google Scholar 

  • Zhang S, Karunramuni RJ (2000) Boundary bias correction for nonparametric deconvolution. Ann Inst Stat Math 52:612–629

    Article  Google Scholar 

Download references

Author information

Authors and Affiliations

  1. School of Management, Huazhong University of Science and Technology, Wuhan, China

    Jun Cai

  2. Department of Economics, Syracuse University, Syracuse, NY, USA

    William C. Horrace

  3. Department of Economics, University of Miami, Miami, FL, USA

    Christopher F. Parmeter

Authors
  1. Jun Cai
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. William C. Horrace
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Christopher F. Parmeter
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding authors

Correspondence to Jun Cai, William C. Horrace or Christopher F. Parmeter.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Cai, J., Horrace, W.C. & Parmeter, C.F. Density deconvolution with Laplace errors and unknown variance. J Prod Anal 56, 103–113 (2021). https://doi.org/10.1007/s11123-021-00612-1

Download citation

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s11123-021-00612-1

Keywords

Search

Navigation