دانلود رایگان مقاله لاتین ترکیب پیش بینی انتخاب میانگین از سایت الزویر
عنوان فارسی مقاله:
زمان برای انتخاب میانگین ساده در ترکیب پیش بینی
عنوان انگلیسی مقاله:
When to choose the simple average in forecast combination
سال انتشار : 2016
برای دانلود رایگان مقاله ترکیب پیش بینی انتخاب میانگین اینجا کلیک نمایید.
بخشی از مقاله انگلیسی:
2. Related work
A substantial amount of research has been conducted on the performance and robustness of SA in comparison to other forecast combination methods. A basic and intuitive finding is that the performance of SA depends on the ratio of the error variances of the forecasts as well as on their correlation. SA can be expected to perform well in case of similar error variances and low or medium error correlations (Bunn, 1985; Gupta & Wilton, 1987), since the weights which are optimal in the evaluation sample then approach equal weights. However, as shown by Dickinson (1973); Winkler and Clemen (1992), and Smith and Wallis (2009), SA can outperform other methods even for differing error variances or strongly correlated errors because of instable weight estimates. Elliott (2011) found that gains from using OW instead of SA are often too small to balance estimation errors. Claeskens, Magnus, Vasnev, and Wang (2016) showed that weight estimation can even introduce biases in combinations of unbiased forecasts. Monte Carlo simulations by Kang (1986) and Gupta and Wilton (1987) confirmed that unstable weight estimates are key to the high competitiveness of SA. Evaluations on real-world data, for instance for U.S. money supply forecasts (Figlewski & Urich, 1983) or GNP forecasts (Kang, 1986; Clemen & Winkler, 1986) showed similar results. Some guidelines to help decision-makers in selecting a combination method have been proposed. In the case of two forecasts, Schmittlein, Kim, and Morrison (1990) recommended SA for small sample sizes and for errors with similar variances and weak correlation. De Menezes, Bunn, and Taylor (2000) recommended SA only for approximately equal error variances and OW for large samples and low error correlation. In other cases, they suggested using outperformance probabilities (with small samples and unequal error variances), optimal weights constrained to the interval [0,1] (with medium or large samples and correlation over 0.5), or OW calculated with a correlation of zero instead of the estimated correlation, i.e., assuming uncorrelated errors (with medium sample sizes and correlations below 0.5). Thresholds for similarity/dissimilarity of error variances and sample size were, however, not quantified. Both guidelines assume equal characteristics (error variances and covariances) of known training and unknown (future) observations. However, these characteristics might change over time because of structural changes in time series, which might influence the performance of OW and SA very differently. Miller, Clemen, and Winkler (1992) showed that SA can, in comparison to OW and other approaches, benefit from several types of structural breaks such as location shifts. Diebold and Pauly (1987) found that structural changes generally tend to impact complex approaches more than simpler ones as the estimated weights tend to increasingly differ from the ones that would minimize error in the evaluation sample. In this paper, in contrast to existing guidelines, we propose an analytical model to determine whether SA will asymptotically outperform OW in a specific setting. We derive decision rules based on statistical considerations that do not only consider sample size and variance/ covariance estimates, but also how much those values are allowed to divergence between training and evaluation sample for a decision to stay optimal. These thresholds are key to assessing the robustness of a decision but have received scant attention in the literature so far.
برای دانلود رایگان مقاله ترکیب پیش بینی انتخاب میانگین اینجا کلیک نمایید.
کلمات کلیدی:
forecasting - Combination Forecast - Which models to pick? - Cross ... https://stats.stackexchange.com/questions/.../combination-forecast-which-models-to-pi... Mar 25, 2015 - I plan to do a combination forecast for real estate cycle prediction. ... find conditions under which you may just use a simple average and when ... On the Forecast Combination Puzzle https://arxiv.org/pdf/1505.00475 by W Qian - 2015 - Related articles May 3, 2015 - It is often reported in forecast combination literature that a simple average of candidate forecasts is more robust than sophisticated combining ... [PDF]A Simple Explanation of the Forecast Combination Puzzle* www.warwick.ac.uk/fac/soc/economics/staff/academic/.../smithwallis_obes_09.pdf by J SMITH - 2008 - Cited by 137 - Related articles average of competing forecasts is expected to be more accurate, in terms of MSFE, than a combination based on estimated weights. The article proceeds as ... [PDF]Package 'ForecastCombinations' https://cran.r-project.org/web/.../ForecastCombinations/ForecastCombinations.pdf Nov 23, 2015 - ers: Simple average, Ordinary Least Squares, Least Absolute Deviation, ... Forecasts combination using regression, robust regression, con-. Searches related to average in forecast combination granger and ramanathan improved methods of combining forecasts how to combine forecasts averaging and the optimal combination of forecasts bates granger 1969 forecast combination r how to combine multiple forecasts forecast combination puzzle
