## Decisionmetrics: a decision-based approach to econometric modelling (2007)

Venue: | Journal of Econometrics |

Citations: | 14 - 0 self |

### BibTeX

@ARTICLE{Skouras07decisionmetrics:a,

author = {Spyros Skouras and Jel C and Doyne Farmer and Søren Johansen and Shareen Joshi and Steve Satchell},

title = {Decisionmetrics: a decision-based approach to econometric modelling},

journal = {Journal of Econometrics},

year = {2007},

pages = {414--440}

}

### OpenURL

### Abstract

In many applications it is necessary to use a simple and therefore highly misspecified econometric model as the basis for decision-making. We propose an approach to developing a possibly misspecified econometric model that will be used as the beliefs of an objective expected utility maximiser. A discrepancy between model and ‘truth ’ is introduced that is interpretable as a measure of the model’s value for this decision-maker. Our decision-based approach utilises this discrepancy in estimation, selection, inference and evaluation of parametric or semiparametric models. The methods proposed nest quasilikelihood methods as a special case that arises when model value is measured by the Kullback-Leibler information discrepancy and also provide an econometric approach for developing parametric decision rules (e.g. technical trading rules) with desirable properties. The approach is illustrated and applied in the context of a CARA investor’s decision problem for which analytical, simulation and empirical results suggest it is very effective.

### Citations

9827 | The Nature of Statistical Learning Theory
- Vapnik
- 1995
(Show Context)
Citation Context ...nalog of the pseudooptimal parameter bc ∗ (ZT )=argmax c∈B TX u (yt,a(g (xt,c))) (28) t=1 does indeed converge to the pseudo-optimal parameter. Since this is an Empirical Risk Minimisation estimator (=-=Vapnik 1998-=-) it may be referred to as a VOID-ERM estimator. Skouras (1998) appears to provide the only application of such a VOID-ERM estimator to date and results reported there suggest that predictive densitie... |

1443 | Generalized Autoregressive Conditional Heteroskedasticity - Bollerslev |

1378 | Statistical Decision Theory and Bayesian Analysis, Second Edition - Berger - 1985 |

1286 | Information theory and statistics - Kullback - 1968 |

1150 | Bayesian theory - Bernardo, Smith - 1994 |

856 | Comparing predictive accuracy
- Diebold, Mariano
- 1995
(Show Context)
Citation Context ...operties (such as a quasi-likelihood function) is familiar to economists primarily from the literature on forecasting and forecast evaluation under general cost-of-error loss functions (Granger 1969, =-=Diebold and Mariano 1995-=-, Pesaran and Skouras 2001), though it has also appeared in the context of estimation (Weiss 1996, Manski 1991). This is conceptually distinct from the literature on robust statistics where the search... |

759 | Specification tests in econometrics - HAUSMAN - 1978 |

536 | Testing statistical hypotheses - Lehmann - 1986 |

479 | Theoretical Statistics - Cox, Hinkley - 1974 |

397 | Model Selection and Inference: a Practical Information-theoretic Approach - Burnham, Anderson - 1998 |

368 |
On measures of entropy and information
- Rényi
- 1961
(Show Context)
Citation Context ...ng: firstly, it has been argued that the pseudo-true model provides the best description of ‘truth’ feasible within the constraints of a misspecified model (Akaike 1985, Larimore 1983, Kerridge 1961, =-=Renyi 1961-=-); secondly, KLID is a very interpretable quantity that plays a central role in statistics in general (e.g. Bernardo and Smith 11By contrast, the subjective Bayesian approach is to construct ‘rational... |

343 | Investing in the Long-run when Returns are Predictable
- Barberis
- 2000
(Show Context)
Citation Context ...dard methods that require working with such models. Inevitably, econometric applications of statistical decision theory in econometrics (for example Bayes decision rules in financial decision-making (=-=Barberis 1999-=-, Bawa, Brown, and Klein 1979) or extensions designed to accommodate model uncertainty (Chamberlain 2000)) have proceeded by ignoring the most important but less tractable types of misspecification (t... |

261 | Simple technical trading rules and the stochastic properties of stock returns - Brock, Lakonishok, et al. - 1991 |

148 | Model Selection - Linhardt, Zucchini - 1986 |

144 | Estimation, Inference and Specification Analysis - White - 1994 |

141 | The Analytics of Uncertainty and Information - Hirshleifer, Riley - 1992 |

138 | Estimating Portfolio and Consumption Choice: A Conditional Euler Equation Approach - Brandt - 1999 |

109 | Transaction costs and predictability: Some utility cost calculations - Balduzzi, Lynch - 1999 |

105 | Approximating Expected Utility by a Function of Mean and Variance - Levy, Markowitz - 1979 |

96 | Variable selection for portfolio choice - Ait-Sahalia, Brandt - 2001 |

93 | Estimation risk and optimal portfolio choice - Bawa, Brown, et al. - 1979 |

89 | Econometric Foundations - Mittelhammer, Judge, et al. - 2000 |

87 |
Economic forecast evaluation: profits versus the conventional error measures
- Leitch, Tanner
- 1991
(Show Context)
Citation Context ...ls is generally sensitive to the objective function with respect to which it is evaluated. This has been established on both theoretical (Geisel and Zellner 1968, Zellner 1973) and empirical grounds (=-=Leitch and Tanner 1991-=-, West, Edison, and Cho 1993) and has consequently led to the literature on decision-based approaches to forecast evaluation (Pesaran and Skouras 2001). It would therefore be remarkable if entropy max... |

77 | Data snooping, technical trading rule performance, and the bootstrap, The - Sullivan, Timmermann, et al. - 1999 |

59 |
Prediction with a Generalized Cost of Error Function
- Granger
- 1969
(Show Context)
Citation Context ...he nature of this class is rather unusual). Conceptually, this class is related to more familiar M-estimators based on two other types of objective functions: ‘relevant’ cost-of-error loss functions (=-=Granger 1969-=-, Diebold and Mariano 1995, Pesaran and Skouras 2001, Weiss 1996, Manski 1991) and quasi-likelihood functions. The use of the first type of objective functions has been proposed consequently to both t... |

57 | Quasi-Likelihood And Its Application - Heyde - 1997 |

52 | 2000), Foundations of Technical Analysis - Lo, Mamaysky, et al. |

43 | If youre so smart, why arent you rich? Belief selection in complete and incomplete markets - Blume, Easley - 2000 |

43 | Robust statistical procedures - Huber - 1997 |

39 | Pitfalls of a Minimax Approach to Model Uncertainty - Sims - 2001 |

39 | A Utility-Based Comparison of Some Models of Exchange Rate - West, Edison, et al. - 1993 |

37 | Econometric Theory - Davidson - 2000 |

34 | Note on the consistency of the maximum likelihood estimate for nonidentifiable distributions - Redner - 1981 |

28 |
Goodness of prediction fit
- Aitchison
- 1975
(Show Context)
Citation Context ...ic modelling (1994), Zellner (1988), Kullback (1968)); thirdly, it is very convenient (it is usually fairly tractable and in some cases we even be used to rank predictive models without knowing fY,X (=-=Aitchison 1975-=-)). 12 However, all these justifications are irrelevant from our decision-maker’s perspective for whom the best feasible description of ‘truth’ will be given by the model with parameter c ∗ ∈ B minimi... |

28 |
Estimating time series models using the relevant cost function
- Weiss
- 1996
(Show Context)
Citation Context ...asting and forecast evaluation under general cost-of-error loss functions (Granger 1969, Diebold and Mariano 1995, Pesaran and Skouras 2001), though it has also appeared in the context of estimation (=-=Weiss 1996-=-, Manski 1991). This is conceptually distinct from the literature on robust statistics where the search is for loss functions that when used in estimation lead to models satisfying some standard notio... |

26 |
Econometrics and Decision Theory
- Chamberlain
- 2000
(Show Context)
Citation Context ...al decision theory in econometrics (for example Bayes decision rules in financial decision-making (Barberis 1999, Bawa, Brown, and Klein 1979) or extensions designed to accommodate model uncertainty (=-=Chamberlain 2000-=-)) have proceeded by ignoring the most important but less tractable types of misspecification (this point is also made by Sims (2001) in describing how model uncertainty has been incorporated in certa... |

25 | Dynamic Portfolio Selection by Augmenting the Asset Space - Brandt, Santa-Clara - 2006 |

16 | Aït-Sahalia (2001): “Variable Selection for Portfolio Choice - Brandt, Y |

15 | Bayesian decision theory and the simplification of models - KADANE, DicKEY - 1980 |

11 |
Prediction and Entropy,” in A Celebration of Statistics
- Akaike
- 1985
(Show Context)
Citation Context ...ns to the pseudo-true model are the following: firstly, it has been argued that the pseudo-true model provides the best description of ‘truth’ feasible within the constraints of a misspecified model (=-=Akaike 1985-=-, Larimore 1983, Kerridge 1961, Renyi 1961); secondly, KLID is a very interpretable quantity that plays a central role in statistics in general (e.g. Bernardo and Smith 11By contrast, the subjective B... |

10 | Probability, Econometrics and Truth - Keuzenkamp - 2000 |

8 | Economic Comparability of - Marschak, Miyasawa - 1968 |

7 | Viceira (1999): “Consumption and Portfolio Decisions When Expected Returns are Time Varying,” Quarterly - Campbell, M |

7 |
Predictive inference, sufficiency, entropy and an asymptotic likelihood principle
- Larimore
- 1983
(Show Context)
Citation Context ...udo-true model are the following: firstly, it has been argued that the pseudo-true model provides the best description of ‘truth’ feasible within the constraints of a misspecified model (Akaike 1985, =-=Larimore 1983-=-, Kerridge 1961, Renyi 1961); secondly, KLID is a very interpretable quantity that plays a central role in statistics in general (e.g. Bernardo and Smith 11By contrast, the subjective Bayesian approac... |

5 | Statistical Modelling of Asymmetric Risk - Knight, Satchell, et al. - 1995 |

4 | Decisions with Estimation Uncertainty - Klein, Rafsky, et al. - 1978 |

2 |
Encyoclopaedia of Statistical Scienceschap
- Burbea
- 1983
(Show Context)
Citation Context ...re distributional discrepancies rather than error loss functions. While a number of distributional discrepancy measures have been considered as potential loss functions in the statistical literature (=-=Burbea 1983-=-), this seems to be the first paper to define a discrepancy as a derivative of an explicit decision-making problem and to therefore provide a compelling reason for using distributional discrepancies a... |

2 |
Sensitivity of Control to Uncertainty and
- Geisel, Zellner
- 1968
(Show Context)
Citation Context ...evious research suggests that the performance of economic models is generally sensitive to the objective function with respect to which it is evaluated. This has been established on both theoretical (=-=Geisel and Zellner 1968-=-, Zellner 1973) and empirical grounds (Leitch and Tanner 1991, West, Edison, and Cho 1993) and has consequently led to the literature on decision-based approaches to forecast evaluation (Pesaran and S... |

2 |
28 A decision-based approach to econometric modelling
- Granger
- 1969
(Show Context)
Citation Context ...ing general properties (such as a quasi-likelihood function) is familiar to economists primarily from the literature on forecasting and forecast evaluation under general cost-of-error loss functions (=-=Granger 1969-=-, Diebold and Mariano 1995, Pesaran and Skouras 2001), though it has also appeared in the context of estimation (Weiss 1996, Manski 1991). This is conceptually distinct from the literature on robust s... |

2 | Risk Neutral Forecasting,” Working paper 98-40 - Skouras - 1998 |