Smooth Move Exponential Smoothing
James W. Taylor
Saïd Business University
University of Oxford
Diary of Forecasting, 2004, Vol. 23, pp. 385-394.
Talk about for Correspondence:
James T. Taylor
Saïd Business School
University of Oxford
Playground End Avenue
Oxford OX1 1HP, UK
Tel: +44 (0)1865 288927
Fax: +44 (0)1865 288805
Email: adam. [email protected] ox. ac. uk
Smooth Move Exponential Smoothing
SMOOTH MOVE EXPONENTIAL SMOOTHING
Adaptive exponential smoothing methods enable a smoothing parameter to alter over time, in order to adapt to changes in the characteristics of the time series. However , these methods have maintained to produce volatile forecasts and also have performed poorly in empirical studies. This paper gives a new adaptive method, which in turn enables a smoothing variable to be modelled as a logistic function of any user-specified variable. The procedure is similar to that accustomed to model the time-varying unbekannte in clean transition types. Using lab-created data, all of us show the fact that new strategy has the potential to outperform existing adaptive methods and constant parameter strategies when the estimation and analysis samples both contain a level shift or perhaps both contain an outlier. An scientific study, making use of the monthly time series from the M3Competition, offered encouraging results for the new approach.
Keywords: Adaptive rapid smoothing; Smooth transition; Level shifts; Outliers
Exponential smoothing is an easy and sensible approach to foretelling of, whereby the forecast is definitely constructed from a great exponentially measured average of past findings. The literary works generally suggests that the smoothing parameters ought to be estimated through the data, generally by minimising the amount of ex lover post 1-step-ahead forecast errors (Gardner, 1985). Some experts have argued that the guidelines should be permitted to change as time passes, in order to adjust to the latest attributes of the time series. For example , in the event there has been a level shift in the series, the exponentially measured average ought to adjust to ensure that an even greater pounds is put on the most recent remark. A variety of adaptive exponential smoothing methods had been developed to deal with this problem. Nevertheless , these strategies have been rebuked for bringing about unstable predictions (e. g. Fildes, 1979), and, indeed, empirical research have shown that they can be less successful than the easier, traditional treatment of continuous optimised parameters (e. g. Makridakis ainsi que al., 1982). This paper presents a new adaptive dramatical smoothing technique, which enables a smoothing parameter to become modelled as a logistic function of a user-specified variable. One simple choice for this variable is a magnitude from the forecast mistake in the previous period. The new method is analogous to this used to version the time-varying parameter in smooth transition models (see Teräsvirta, 1998). Using lab-created and real data, we show that the new adaptive method problems the overall performance of set up adaptive techniques and frequent parameter methods.
In the next section, we review the literature on adaptive exponential smoothing methods. We then bring in our new approach. The fourth section demonstrates the potential of the approach employing two lab-created series. Inside the fifth section, we execute a large-scale simulation study in order to compare the forecast precision of the fresh method get back of additional methods. We all then make use of a large info set of real-time series to evaluate the strength of the fresh method. The ultimate section supplies a summary and concluding remarks. 1
ESTABLISHED ADAPTIVE EXPONENTIAL SMOOTHING METHODS
There have been many different proposals to get enabling the exponential smoothing parameters to adapt after some time according to the characteristics of the series. Williams (1987) writes that it can be widely acknowledged that in multi-dimensional smoothing, such as the Holt and Holt-Winters methods, only the smoothing parameter for the...
References: Adya M, Collopy F, Armstrong JS, Kennedy M. 2001. Automatic id of time series
features pertaining to rule-based forecasting
Balke NATURSEKT. 1993. Finding level adjustments in time series. Journal of Business and Economic
Stats 11: 81-92.
Bunn DW. 1981. Adaptive forecasting making use of the Kalman filter. Omega being unfaithful: 323-324.
Dennis JD. 78. A efficiency test of a run-based adaptive exponential smoothing.
Ekern S i9000. 1981. Adaptable exponential smoothing revisited. Record of the Functional
Research Society 32: 775-782.
Ekern T. 1982. Upon simulation studies of adaptive forecasts. Omega 10: 91-93.
Enns PG, Machak JA, Spivey CALIFORNIA, Wroblesk WJ. 1982. Predicting applications of an adaptive
multiple exponential smoothing model
Fildes R. lates 1970s. Quantitative forecasting – your the fine art: extrapolative designs. Journal of
the Operational Research Contemporary society 30: 691-710.
Gardner SERA Jr. 85. Exponential smoothing: the state of the art. Diary of Forecasting 4: 128.
Gardner FUE Jr, Dannenbring DG. 1980. Forecasting with exponential smoothing: Some
suggestions for unit selection
Granger CWJ, Teräsvirta T. 1993. Modelling non-linear Economic Relationships, Advanced
Text messages in Econometrics, Oxford University Press: Ny.
Harvey AC. 1990. Foretelling of, structural period series types and the Kalman filter. Cambridge
University Press: New York.
Hyndman RJ, Koehler AB, Ord JK, Snyder RD. 2001. Prediction intervals for exponential
smoothing express space types
Hyndman RJ, Koehler STOMACH, Snyder RD, Grose S i9000. 2002. A situation space construction for programmed
forecasting employing exponential smoothing methods
Makridakis S, Hibon M. 2000. The M3-Competition: results, a conclusion and implications.
Mentzer JT. 1988. Predicting with adaptable extended rapid smoothing. Diary of the
Schools of Marketing Research 16: 62-70.
Mentzer JT, Gomes, 3rd there�s r. 1994. Further extensions of adaptive prolonged exponential smoothing
and assessment with the M-Competition
Pantazopoulos SN, Pappis CP. 1996. A fresh adaptive way of extrapolative predicting
Snyder RD. 1988. Progressive Fine tuning of Simple Exponential Smoothing Forecasts. Diary
of the Detailed Research Society 39: 393-399.
Taylor JW, Bunn DW. 1999. A quantile regression approach to creating prediction intervals.
Taylor JW. 2003a. Damped multiplicative pattern exponential smoothing. Forthcoming
Foreign Journal of Forecasting.
Taylor JW. 2003b. Volatility forecasting with clean transition dramatical smoothing.
Trigg DW, Leach AG. 1967. Exponential smoothing with an adaptive response rate.
Tsay RS. 1988. Outliers, level shifts, and variance changes in time series. Journal of
Forecasting six: 1-20.
vehicle Dobben para Bruyn CS. 1964. Conjecture by progressive correction. Journal of the Regal
Statistical Contemporary society Series W 26: 113-122.
Whybark DC. 1973. Comparison of adaptive forecasting techniques. Logistics Transportation
Review 8: 13-26.
Williams TM. 1987. Adaptable Holt-Winters foretelling of. Journal in the Operational Analysis
Society 38: 553-560.
Willaims DW, Miller D. 99. Level-adjusted dramatical smoothing intended for modelling prepared
Small PC. 99. Nonstationary period series evaluation and forecasting. Progress in
Environmental Scientific research 1: 3-48.
Young PERSONAL COMPUTER, Pedregal DISC JOCKEY, Tych T. 1999. Active harmonic regression. Journal of
Forecasting 18: 369-394.
Young PC. the year 2003. Identification of your time varying systems. Forthcoming in Encyclopaedia of
Life Support Systems, UNESCO.