Abstract
The problem of prediction is revisited with a view towards going beyond the typical nonparametric setting and reaching a fully model-free environment for predictive inference, i.e., point predictors and predictive intervals. A basic principle of model-free prediction is laid out based on the notion of transforming a given setup into one that is easier to work with, namely i.i.d. or Gaussian. As an application, the problem of nonparametric regression is addressed in detail; the model-free predictors are worked out, and shown to be applicable under minimal assumptions. Interestingly, model-free prediction in regression is a totally automatic technique that does not necessitate the search for an optimal data transformation before model fitting. The resulting model-free predictive distributions and intervals are compared to their corresponding model-based analogs, and the use of cross-validation is extensively discussed. As an aside, improved prediction intervals in linear regression are also obtained.
Similar content being viewed by others
Notes
The qualitative difference is that the interest of the MF practitioner is on observable quantities, i.e., current and future data, as opposed to unobservable model parameters and estimates thereof. In this sense, despite being frequentist in nature, the MF principle is in concordance with Bruno de Finetti’s statistical philosophy—see e.g. Dawid (2004) and the references therein.
Rather than doing a two-dimensional search over h and q to minimize PRESS, the simple constraint q=h will be imposed in what follows which has the additional advantage of rendering \(M_{x}\geq m^{2}_{x}\) as needed for a well-defined estimator \(s^{2}_{x}\) in Eq. (11). Note that the choice q=h is not necessarily optimal; see e.g. Wang et al. (2008). Furthermore, note that these are global bandwidths; techniques for picking local bandwidths, i.e., a different optimal bandwidth for each x, are widely available but will not be discussed further here in order not to obscure the paper’s main focus. Similarly, there are several recent variations on the cross-validation theme such as the one-sided cross-validation of Hart and Yi (1998), and the far casting cross-validation for dependent data of Carmack et al. (2009) that present attractive alternatives. However, our discussion will focus on the well-known standard form of cross-validation for concreteness especially since our aim is to show how the Model-Free prediction principle applies in nonparametric regression with any type of kernel smoother, and any type of bandwidth selector.
Here, and for the remainder of Sect. 3, we will assume that the form of the estimator m x is linear in the Y data; our running example of a kernel smoother obviously satisfies this requirement, and so do other popular methods such as local polynomial fitting.
Strictly speaking, the W t ’s are not exactly independent because of dependence of \(m_{x_{t}}\) and \(s_{x_{t}}\) to \(m_{x_{k}}\) and \(s_{x_{k}}\). However, under typical conditions, \(m_{x}\stackrel{P}{\longrightarrow}E(Y|x)\) and \(s^{2}_{x}\stackrel {P}{\longrightarrow}\mathit{Var}(Y|x)\) as n→∞. Therefore, the W t ’s are—at least—asymptotically independent.
If σ 2(x) is not assumed constant, then \(\tilde{e}_{t}= e_{t} C_{t}/(1-\delta_{x_{t}})\) where \(C_{t}=s_{x_{t}}/s_{x_{t}}^{(t)}\).
For \(\bar{D}_{x_{\mathrm{f}}}^{-1}\) to be an accurate estimator of \(D_{x_{\mathrm{f}}}^{-1}\), the value x f must be such that it has an appreciable number of h-close neighbors among the original predictors x 1,…,x n as discussed in Remark 4.1. As an extreme example, note that prediction of Y f when x f is outside the range of the original predictors x 1,…,x n , i.e., extrapolation, is not feasible in the model-free paradigm.
References
Altman NS (1992) An introduction to kernel and nearest-neighbor nonparametric regression. Am Stat 46(3):175–185
Atkinson AC (1985) Plots, transformations and regression. Clarendon, Oxford
Beran R (1990) Calibrating prediction regions. J Am Stat Assoc 85:715–723
Bickel P, Li B (2006) Regularization in statistics. Test 15(2):271–344
Box GEP, Cox DR (1964) An analysis of transformations. J R Stat Soc, Ser B, Stat Methodol 26:211–252
Breiman L, Friedman J (1985) Estimating optimal transformations for multiple regression and correlation. J Am Stat Assoc 80:580–597
Carmack PS, Schucany WR, Spence JS, Gunst RF, Lin Q, Haley RW (2009) Far casting cross-validation. J Comput Graph Stat 18(4):879–893
Carroll RJ, Ruppert D (1988) Transformations and weighting in regression. Chapman & Hall, New York
Carroll RJ, Ruppert D (1991) Prediction and tolerance intervals with transformation and/or weighting. Technometrics 33:197–210
Cox DR (1975) Prediction intervals and empirical Bayes confidence intervals. In: Gani J (ed) Perspectives in probability and statistics. Academic Press, London, pp 47–55
Dai J, Sperlich S (2010) Simple and effective boundary correction for kernel densities and regression with an application to the world income and Engel curve estimation. Comput Stat Data Anal 54(11):2487–2497
DasGupta A (2008) Asymptotic theory of statistics and probability. Springer, New York
Davison AC, Hinkley DV (1997) Bootstrap methods and their applications. Cambridge University Press, Cambridge
Dawid AP (2004) Probability, causality, and the empirical world: a Bayes–de Finetti–Popper–Borel synthesis. Stat Sci 19(1):44–57
Draper NR, Smith H (1998) Applied regression analysis, 3rd edn. Wiley, New York
Efron B (1979) Bootstrap methods: another look at the jackknife. Ann Stat 7:1–26
Efron B (1983) Estimating the error rate of a prediction rule: improvement on cross-validation. J Am Stat Assoc 78:316–331
Efron B, Tibshirani RJ (1993) An introduction to the bootstrap. Chapman & Hall, New York
Fan J, Gijbels I (1996) Local polynomial modelling and its applications. Chapman & Hall, London
Freedman DA (1981) Bootstrapping regression models. Ann Stat 9:1218–1228
Gangopadhyay AK, Sen PK (1990) Bootstrap confidence intervals for conditional quantile functions. Sankhya, Ser A 52(3):346–363
Goldberger AS (1962) Best linear unbiased prediction in the generalized linear regression model. J Am Stat Assoc 57:369–375
Geisser S (1993) Predictive inference: an introduction. Chapman & Hall, New York
Hahn J (1995) Bootstrapping quantile regression estimators. Econom Theory 11(1):105–121
Hall P (1992) The bootstrap and edgeworth expansion. Springer, New York
Hall P (1993) On edgeworth expansion and bootstrap confidence bands in nonparametric curve estimation. J R Stat Soc, Ser B, Stat Methodol 55:291–304
Hall P, Wehrly TE (1991) A geometrical method for removing edge effects from kernel type nonparametric regression estimators. J Am Stat Assoc 86:665–672
Härdle W (1990) Applied nonparametric regression. Cambridge University Press, Cambridge
Härdle W, Bowman AW (1988) Bootstrapping in nonparametric regression: local adaptive smoothing and confidence bands. J Am Stat Assoc 83:102–110
Härdle W, Marron JS (1991) Bootstrap simultaneous error bars for nonparametric regression. Ann Stat 19:778–796
Hart JD (1997) Nonparametric smoothing and lack-of-fit tests. Springer, New York
Hart JD, Yi S (1998) One-sided cross-validation. J Am Stat Assoc 93(442):620–631
Hong Y (1999) Hypothesis testing in time series via the empirical characteristic function: a generalized spectral density approach. J Am Stat Assoc 94:1201–1220
Hong Y, White H (2005) Asymptotic distribution theory for nonparametric entropy measures of serial dependence. Econometrica 73(3):837–901
Horowitz J (1998) Bootstrap methods for median regression models. Econometrica 66(6):1327–1351
Koenker R (2005) Quantile regression. Cambridge University Press, Cambridge
Li Q, Racine JS (2007) Nonparametric econometrics. Princeton University Press, Princeton
Linton OB, Sperlich S, van Keilegom I (2008) Estimation of a semiparametric transformation model. Ann Stat 36(2):686–718
Loader C (1999) Local regression and likelihood. Springer, New York
McCullagh P, Nelder J (1983) Generalized linear models. Chapman & Hall, London
McMurry T, Politis DN (2008) Bootstrap confidence intervals in nonparametric regression with built-in bias correction. Stat Probab Lett 78:2463–2469
McMurry T, Politis DN (2010) Banded and tapered estimates of autocovariance matrices and the linear process bootstrap. J Time Ser Anal 31:471–482
Nadaraya EA (1964) On estimating regression. Theory Probab Appl 9:141–142
Neumann M, Polzehl J (1998) Simultaneous bootstrap confidence bands in nonparametric regression. J Nonparametr Stat 9:307–333
Olive DJ (2007) Prediction intervals for regression models. Comput Stat Data Anal 51:3115–3122
Pagan A, Ullah A (1999) Nonparametric econometrics. Cambridge University Press, Cambridge
Patel JK (1989) Prediction intervals: a review. Commun Stat, Theory Methods 18:2393–2465
Politis DN (2003) A normalizing and variance-stabilizing transformation for financial time series. In: Akritas MG, Politis DN (eds) Recent advances and trends in nonparametric statistics. Elsevier, Amsterdam, pp 335–347
Politis DN (2007a) Model-free vs. model-based volatility prediction. J Financ Econom 5(3):358–389
Politis DN (2007b) Model-free prediction. In: Bulletin of the international statistical institute—volume LXII, Lisbon, 22–29 Aug 2007, pp 1391–1397
Politis DN (2010) Model-free model-fitting and predictive distributions. Discussion Paper, Department of Economics, Univ of California—San Diego. Retrieved from: http://escholarship.org/uc/item/67j6s174
Politis DN, Romano JP, Wolf M (1999) Subsampling. Springer, New York
Rosenblatt M (1952) Remarks on a multivariate transformation. Ann Math Stat 23:470–472
Ruppert D, Cline DH (1994) Bias reduction in kernel density estimation by smoothed empirical transformations. Ann Stat 22:185–210
Schmoyer RL (1992) Asymptotically valid prediction intervals for linear models. Technometrics 34:399–408
Schucany WR (2004) Kernel smoothers: an overview of curve estimators for the first graduate course in nonparametric statistics. Stat Sci 19:663–675
Seber GAF, Lee AJ (2003) Linear regression analysis. Wiley, New York
Shao J, Tu D (1995) The jackknife and bootstrap. Springer, New York
Shi SG (1991) Local bootstrap. Ann Inst Stat Math 43:667–676
Stine RA (1985) Bootstrap prediction intervals for regression. J Am Stat Assoc 80:1026–1031
Tibshirani R (1988) Estimating transformations for regression via additivity and variance stabilization. J Am Stat Assoc 83:394–405
Tibshirani R (1996) Regression shrinkage and selection via the Lasso. J R Stat Soc, Ser B, Stat Methodol 58(1):267–288
Wang L, Brown LD, Cai TT, Levine M (2008) Effect of mean on variance function estimation in nonparametric regression. Ann Stat 36:646–664
Watson GS (1964) Smooth regression analysis. Sankhya, Ser A 26:359–372
Wolfowitz J (1957) The minimum distance method. Ann Math Stat 28:75–88
Acknowledgements
A preliminary version of this paper was presented as a Plenary Talk at the 10th International Vilnius Conference on Probability and Mathematical Statistics, June 28–July 3, 2010, and as a Special Invited Talk at the 28th European Meeting of Statisticians, August 17–22, 2010; the author is grateful to the audiences in these two—and several other—occasions for their helpful feedback. Many thanks are due to Arthur Berg, Wilson Cheung and Tim McMurry for invaluable help with R functions and computing, and to Richard Davis, Jeff Racine, Bill Schucany, Dimitrios Thomakos and Slava Vasiliev for helpful discussions. The author is also grateful to the Editors, Ricardo Cao and Domingo Morales, for their support and encouragement, and to six (!) anonymous referees for their very detailed and constructive comments; one of the referees deserves special thanks for an astute observation that helped shed light on the workings of the ‘uniformize’ algorithm of Sect. 4. This work has been partially supported by NSF grants DMS-07-06732 and DMS-10-07513, and by a fellowship from the Guggenheim Foundation.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Politis, D.N. Model-free model-fitting and predictive distributions. TEST 22, 183–221 (2013). https://doi.org/10.1007/s11749-013-0317-7
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11749-013-0317-7
Keywords
- Bootstrap
- Cross-validation
- Frequentist prediction
- Heteroskedasticity
- Nonparametric estimation
- Prediction intervals
- Regression
- Smoothing
- Transformations