Modeling of economic forecasts
2016 December 19th
14:46
1. Forecasting methods and selection of predictive models
A. Selecting a predictive analysis method
(1) Classification of economic forecasting methods
Qualitative analysis: For the judgment of the objective object which is difficult to measure by data and information, the qualitative analysis methods commonly used in economic research include expert evaluation Method (Delphi method), Judgment forecast method, market investigation method, analogy method, etc.
Quantitative analysis: Focusing on the use of statistical and measurement of analytical methods, the data collected for processing, so as to obtain the required data results. The quantitative analysis methods of economic forecast include survey forecast, correlation and regression prediction, trend forecast, seasonal forecast, input-output forecast, Markov prediction, production function forecast, short-term forecast, medium and long-term prediction, prediction, extension prediction, causality prediction, etc.
B. Selection of predictive indicators and evaluation system for determining indicators
(1) forecast index and Index system:
Index system refers to an evaluation index system which can comprehensively reflect the shape, nature and change of many elements of economic operation, and the research object is comprehensive and
In-depth, objective evaluation and analysis. It is composed of a series of indicators. It is important to note that the data obtained from the most primitive or the indicators with objective data is called the lowest index. And by the combination of its calculation and other indicators can be composed of a higher level of indicators (not rigorous can be considered to be at different levels of the level of the index of the bucket). In general, the higher the level of the number of indicators less.
(2) indicator selection
The indicator selection first according to the requirements of the indicator calculation, of course, according to the indicator itself changes in the trend and characteristics:
Some indicators have obvious trend of change, can be directly used in the prediction of various models. And some of the indicators are almost variable (example: Per capita land area this indicator, because the land area is basically unchanged, so directly to the population of the prediction on it.) )
C. Selection of evaluation models and construction of predictive models
(1) evaluation Model
(2) predictive models
The main predictive models are linear regression model, moving average model, exponential smoothing model, trend extrapolation model,Arima Prediction model, Markov Prediction model, input-output prediction model, grey Prediction model, artificial neural network prediction model, etc.
(3) Classification of indicators
Different types of indicator data reflect a different level of competitiveness, but if each indicator data is a category, then the analysis is too complex. It can usually be divided into stock, increment, mean, account ratio, ratio (difference for example, growth rate) five kinds.
(4) steps to build a predictive model:
According to the criterion of accuracy, stability and simplicity, the two prediction types of quantitative prediction and qualitative prediction are defined rationally. Secondly, various predictive models are used to analyze the adaptability of all the indicators, and the basis principle and application scope of various predictive models are shown in table 1. The above indexes are forecasted and analyzed by using the historical data. Then, the combined forecasting method is used to synthesize the prediction results, and a more scientific and reasonable forecasting system is formed. Finally, the prediction model is tested, the unreasonable model is corrected, the forecast value and the analysis result are calculated, the forecast error is analyzed, the forecast result is evaluated, and the forecast result is corrected according to the latest feedback information, which makes the prediction result more flexible and maneuverable. The specific steps are shown in Figure 1:
2. Selection experiment and model construction of predictive models
A. Selection criteria for predictive models
(1) selection criteria
According to the predicted objects and their characteristics, the author puts forward the corresponding optimal criteria, such as: the sum of squared error, the smallest error, the smallest relative error, the maximum deviation
The smallest difference.
B. Selection experiment of predictive dominant model
(1) Single indicator leading model
All the indicators in the indicator system are forecasted, according to the actual forecast results, the evaluation criteria of 5 prediction effects are comprehensively examined, and the best adaptive dominant model is found for each index.
Note: The adaptability is not good, or the prediction effect is slightly worse than the predictive model as an auxiliary model and validation model.
(2) The dominant model of type index prediction
There are two methods of selection: First, the dominant model with the largest number of indicators in a certain class of indicators, as the dominant model of such indicators, abbreviated type number optimal dominant model. The second is the best dominant model of the average prediction effect in some kind of index, as the dominant model of this kind of index, abbreviation type mean optimal dominant model.
It is important to note that :
Using the average method to calculate the values of various evaluation criteria can only reflect a medium level, but not reflect the characteristics and laws of each index itself, ignoring the differences between the indicators. In particular, arithmetic averages are susceptible to extreme values, and individual larger values can easily raise the entire average, making it impossible to measure the true level of the predicted effect with an average.
C. Selection and determination of predictive guidance models
Although we have found out the most adaptable model of each indicator, we have also found the five major types of indicators of the respective predictive leading model, but the same kind of indicators of the different indicators of the trend of the performance is not the same, can not fully use a model of a certain type of all indicators to predict. Therefore, after determining the dominant model of type indicator prediction, we also need to select other predictive models for the few indexes that are not fit for the dominant model, and choose other more adaptable models, which can be called auxiliary models.
3. Application of Combinatorial forecasting model
A. The concept of a combined predictive model
The combination forecasting method is to synthesize the prediction results obtained by multiple forecasting methods by establishing a combined forecasting model, in order to obtain a narrow range of prediction values for the system analysis and decision-making use. The method of linear programming is used to determine the weights of each predictive model, which can automatically select the forecasting model which is most suitable for the trend of indicator change, effectively reduce the impact of a single forecast model by stochastic factors, and make use of various predictive sample information more systematically and comprehensively than single prediction model, thus improving the accuracy and stability of prediction.
Where: the weight of the choice method:
Note:Cij can be interpreted as a sum of squares of all periods of predicted error values in a given method (presumably i=j, because for all I there is always a J equal to it).
Note:S.T said: Make ... Meet ...
5. Evaluation of the effectiveness of predictive evaluation results
A. Accuracy analysis
B. Stability analysis
References: [1] Huangmauchen, Li Ying. The selection experiment and effect analysis of the forecasting model of comprehensive competitiveness of provincial economy [J]. Intelligence Magazine, 2012 (7): 1-9.
Predictive Modeling Step Analysis 1