【作者】 李干琼；

【导师】 许世卫；

【作者基本信息】 中国农业科学院 ， 农业经济管理， 2012， 博士

【摘要】 农产品市场价格事关民众生计和社会稳定。近年来中央一号文件和政府工作报告中多次提到要加强农产品市场监测预警工作，避免农产品市场价格大起大落。农产品市场短期价格受多种复杂因素的影响，使得短期波动加剧，市场短期风险凸显，价格预测难度加大。在我国当前市场主体尚不成熟、市场体系尚不健全、法制环境尚不完善等现状下，农业生产经营者由于难以对市场供求和价格变化做出准确预期，时常要面临和承担价格波动所带来的市场风险；农业行政管理部门也常常因缺少有效的市场价格短期走势的预判信息，难以采取有预见性的事前调控措施；消费者由于缺少权威信息的及时引导，在市场价格频繁波动中极易产生恐慌心理，从而加速价格波动的恶性循环。因此，开展农产品市场短期预测研究，对促进农业生产稳定、农民增收和农产品市场有效供给具有重要的理论和现实意义。本研究选择粮食、畜产品、蔬菜、水果等主要农产品为研究对象，以现代西方经济学、计量经济学、统计学、价格学等有关理论为指导，采用了非参数核密度估计法、H-P滤波法、基于ARCH类模型的风险价值法（VaR）、事件分析法（Event Study）、分位数回归技术、多层前馈神经网络技术、灰色模型法、向量自回归法（VAR）、非平稳时间序列法（SARIMA模型）、季节调整法（CencusX12法）、Holt-Winters法、组合预测法等12种研究方法，深入剖析影响农产品市场价格的强波动类因子（简称S因子）和易波动类因子（简称V因子），研究SV因子交织下的农产品市场价格波动规律，重点突破农产品市场短期预测技术，建立了基于不同信息、不同技术、不同目标的农产品市场价格短期预测模型，初步设计与实现了农产品市场价格智能预测系统。本研究主要开展了以下几个方面的创新性研究：（1）针对农产品市场价格影响因素复杂多样，提出了SV因子的框架分析法思想，开展了SV因子分析法的理论基础研究。本研究将农产品市场价格影响因子划分为强波动因子和易波动类因子，为异常市场波动早期预警提供了新视角；探讨了预测方法的五个发展阶段、农产品市场预测与相关学科的关系和农产品市场价格波动理论等内容。（2）开展了SV因子交织下的农产品市场价格波动研究。一是立足国内视角的常规经济因素和非经济性因素，以蔬菜为例的实证研究得出结论：Granger检验表明生产成本（化肥）、流动性过剩、热钱和气候等因素是蔬菜市场价格波动的原因。化肥价格的影响存在一定滞后期，长期协整向量为[1,-1.18]。流动性过剩（货币发行量）对蔬菜价格短期波动影响较为明显，货币发行量增长1%，蔬菜价格将上涨1.29%；其次为气候（气温）对蔬菜价格短期波动的影响，气温下降或上升1%，蔬菜价格将上涨或下跌0.08%；热钱对蔬菜短期价格波动影响很小。二是立足国际视角的农产品市场价格传导关系，以玉米和大豆为例的实证结果表明：Granger检验表明国际市场价格变动是国内市场价格变动的原因，反之则不成立，且国内玉米价格受国际市场的影响要小于国际大豆价格对国内大豆价格的影响；长期波动关系看，国内玉米价格与国际玉米价格的协整向量为[1,-0.61]，国内大豆价格与国际大豆价格的协整向量为[1,-0.86]；短期波动关系看，上一期国际玉米价格上涨1%，本期国内玉米价格将上涨0.07%，而国际大豆市场价格对国内大豆市场价格的影响是同步的，国际大豆当期价格上涨1%，国内大豆价格将上涨0.44%。（3）采用非参数核密度技术估计了11种蔬菜和12种水果市场收益率的概率密度分布，建立了基于ARCH类模型的农产品市场短期风险动态评估模型。采用核密度技术的估计结果表明：传统意义上的正态分布、Beta分布、Burr分布、Gamma分布等经验分布不是最优的；蔬菜和水果市场收益率的概率密度分布是不对称的，涨价风险要高于降价风险，且大多数蔬菜品种和水果品种的价格大起大落成为常态。以大豆和蔬菜为例，利用GARCH模型分别计算了大豆期货市场收益率的日度VaR和菠菜批发市场收益率的日度VaR、周度VaR、月度VaR，实证分析结果表明：基于GARCH模型计算的风险价值可以较好地反映市场收益率的分布和波动性两个因素，不仅刻画了市场收益率变化的动态过程，适应风险价值的需要，而且在很大程度上也提高了VaR的精度。（4）开展了基于S因子的农产品市场短期建模研究。针对突发性事件主导下的价格预测，在比较分析经典的事件分析法、基于虚拟变量的事件分析法和基于知识的事件分析法等三类方法的基础上，创新性的提出了突发事件的IPAD预测法框架，将事件分解为影响强度（Intensity）、持续时间（Persistence）、衰减模式（Attenuation model）和冲击方向（Direction）等4个要素。针对市场价格的非线性变化和不确定性，建立了多层前馈人工神经网络预测模型，研究结果表明对猪肉、鸡肉和鸡蛋的预测准确性在95%以上，对蔬菜和水果的预测准确性在90%以上。针对数据样本量小的问题，建立了灰色预测GM(1,1)模型，研究结果表明对猪肉、鸡肉和鸡蛋的预测准确性基本都在95%以上。（5）开展了基于V因子的农产品市场短期建模研究。在结构化模型研究方面，分别建立了均值回归预测模型和分位数回归预测模型，并对猪肉、鸡肉和鸡蛋等3种农产品进行了实证分析，预测结果表明：基于均值回归模型的猪肉价格预测精度在90%左右，鸡肉和鸡蛋价格的预测精度在98%以上；基于分位数回归预测模型的预测精度都在99%以上。在时序分析模型研究方面，建立了向量误差修正模型（VEC）、SARIMA模型、Holt-Winters季节指数平滑模型和Census X12季节分解法等4类模型，并对猪肉、鸡肉、鸡蛋、蔬菜和水果等5类农产品进行了实证分析，预测结果表明：4类时间序列模型预测精度都在90%以上，其中鸡蛋和鸡肉的模型预测精度在95%以上。（6）开展了基于过程的模型集成预测和基于预测结果的组合预测研究。由于不同模型对数据信息的提取程度不尽相同，随着时间推移单项预测的稳健性会有所降低，在未来不确定性增多的情况下组合预测往往更具有优势。以蔬菜和水果为例的组合预测研究结果表明：大多数情况下通过组合预测可以提高预测精度，但也存在组合预测效果不如某单一预测方法的样本现象；对预测误差较大的单项预测方法进行组合，预测精度提高的幅度会更大一些，如果进行组合的各种单项预测方法本身的精度都很高，则组合预测精度提高的幅度就很有限。更多还原

【Abstract】 Agricultural product market price is relevant to the livelihood of the people and social stability. Inrecent years, the Central No.1Document and Government Report had several times mentioned tostrengthen agricultural product market Monitoring and early warning so as to ensure that the agriculturalproduct market price does not fluctuate wildly. The price of agricultural product short-term market isinfluenced by multiple complicated factors, which makes short-term fluctuation drastic, short-termmarket risk highlighted, and the difficulty of price forecast increased. Under the situation in whichmarket main body is not mature, market system is not perfect, and legal system environment still is notperfected, etc., agricultural producers often face and take the market risk brought by price fluctuationdue to the difficulty of precise prediction for market supply-demand and price fluctuation. Departmentsof agricultural administration are often short of effective Anticipation information of market priceshort-term trend, and difficult to adopt forward-looking and beforehand control measures. Due to theshortage of timely authority information guide, the consumers are susceptible to panic psychologyduring frequent market prices fluctuation, which hence speeds up the vicious circle of price fluctuation.Therefore, it is of important theoretical and practical significance for us to carry out short-term forecastof agricultural product market, which will help to promote stable agricultural production, increasefarmers’ income, and guarantee the supply of agricultural product market.This dissertation has selected pork, chicken, egg, vegetable and fruit as the research object, takenmodern western economics, econometrics, statistics, price theory and other relevant theories as theguide, and adopted12kinds of study methods of non-parameter kernel density estimation, H-P filteringmethod, value at risk based on ARCH Model, event study, quantile regression, multilayer feed-forwardneural network, grey model method, vector auto-regression method, non-stationary time series method(SARIMA Model), season adjustment method (Cencus X12method), Holt-Winters method, andcombination forecasting method. This study has further analyzed the strong fluctuation class factor (Sfactor) and volatile class factor (V factor) that influence agricultural product market price, studied thefluctuation laws of agricultural product market price intertwined by SV factors, focused onbreakthroughs of short-term forecast technology of agricultural product market, constructed short-termprediction models of agricultural product market price based on different information, differenttechnology, and different objectives, and preliminary designed and realized the intelligent forecastingsystem of agricultural product market price. This study has mainly unfolded innovative studies in thefollowing aspects:(1) To counter the complicated and diversified factors influencing agricultural product market price,this study had put forward the thoughts of framework analysis based on SV factors, and implementedstudies on basic theory of SV factors. This study has broken down the factors influencing agriculturalproduct market price into strong fluctuation class factors and volatile class factors, which has provided anew viewing angle for early warning of abnormal market price fluctuation. In addition, it also exploredthe5development phases of forecasting methods, the relations between agricultural product market forecast and relevant disciplined, as well as fluctuation theories of agricultural product market price andother contents.(2) We have done some research on price fluctuation of agricultural product market interwoven bySV factors. Based on the conventional economic factors and non-economic factors from the perspectiveof China, we have reached the conclusion taking vegetable as the empirical study: it shows productioncost (chemical fertilizer), excess liquidity, hot money and climate are the reasons for vegetable marketprice fluctuation by Granger test. The impacts of chemical fertilizer has certain lag phase and thelong-term co-integration vector is [1,-1.18]. The short-term impact of excess liquidity (amount ofcurrency issue) on vegetable price is fairly significant. When the amount of currency issue increases1%,vegetable price rises by1.29%. Second is the impact of climate (temperature) on vegetable priceshort-term fluctuation. When temperature declines or rises by1%, the price of vegetable will rise ordecline0.08%. The short-term impact of hot money on vegetable price is the minimum. Based on thetransmission relationship from the perspective of agricultural product prices of international market,taking corn and soybeans as an example the empirical study result shows: international market pricechange is the reason for domestic market price change, not vice versa by Granger test. The impact ofinternational corn price on domestic corn price is less than that of international soybean price ondomestic soybean price. Viewing the long-term fluctuation relation, the co-integrated vector of domesticcorn price and international corn price is [1,-0.61] and the co-integrated vector of domestic soybeanprice and international soybean price is [1,-0.86]. Viewing the short-term fluctuation relation, the cornprice of prophase rose by1%while domestic corn price of this phase rises0.07%. However, the impactof international soybean market price on domestic soybean market price is synchronous. Wheninternational soybean price of this phase rises1%, the domestic soybean price will rise0.44%.(3) We have adopted non-parameter kernel density technology to fit the distribution of probabilitydensity function (PDF) of11vegetables and12fruits market returns, and constructed the dynamicassessment model of agricultural product market short-term risk base on ARCH model. The results ofkernel density estimation show: the normal distribution, Beta distribution, Burr distribution, Gammadistribution and other empirical distribution are not the best. The probability density distribution ofvegetable and fruit market returns is asymmetric. The risk of price rising is greater than that of pricedeclining, and the phenomenon of prices wildly fluctuation of vegetable and fruit becomes norm.Taking soybean and vegetable as an example, we have used GARCH model to calculate the daily VaR,weekly VaR and monthly VaR of the market returns. The results of empirical approach show: VaRbased on GARCH model can fairly well reflect the distribution and fluctuation of market returns. It cannot only describe the dynamic course of market returns changes, getting adapted to the needs of VaR,but also to a large extent improve the precision of VaR.(4) We have unfolded short term model construction research on agricultural product market basedon S factors. To counter price forecast guided by unexpected events, on the basis of Comparison ofclassic events study method, event study based on dummy variable and event study based on knowledge,we have innovatively put forward IPAD forecasting framework of unexpected events and divided events into4essential elements of intensity, persistence, attenuation model, and impact direction. Tocounter nonlinear change and uncertainty of market price, we have constructed multilayer feed-forwardneural network model. The results show that the precision of pork, chicken and egg forecast is over95%,and the precision of vegetables and fruits forecast is over90%. To counter the problem of small datasample, we have constructed grey forecasting model GM (1,1). The results show that the forecastprecision of pork, chicken meat and egg basically all over95%.(5) We have unfolded short-term model construction research on agricultural product market basedon V factors. In the aspect of structural model, we have constructed the mean regression forecast modeland the quantile regression forecast model respectively, and conducted empirical analysis of pork,chicken and egg. The forecasting results show that the precision for pork price forecast is about90%and the precision for chicken and egg price forecasts is over98%based on mean regression model, andthe precision of quantile regression forecast model is over99%. In the aspect of time series model, wehave constructed the vector error correction model (VEC), SARIMA model, Holt-Winters seasonexponential smoothing model and Census X12season decomposition method. We have undertakenempirical forecasts of five kinds of agricultural products including pork, chicken meat, egg, vegetableand fruit. The results show that the precision of4kinds of time series models are all over90%, of which,the precision of egg and chicken model is over95%.(6) We have unfolded combination forecasts based on forecasting course and forecasting resultsrespectively. Different models have different extraction degree of information. As time goes on theprecision of individual forecast will be not robust, under the situation of increased uncertainty in thefuture, combination forecast often has more advantages. The results of combination forecasts takingvegetable and fruit as an example show that combination forecasts can improve prediction accuracy inmost cases. However, there are also some special exceptions, i.e. combination forecasting result is notas good as single method. When we undertake combination forecasts based on single method that hasgreater forecasting error, the improvement of prediction precision will be greater. If the precision ofindividual methods is very high, the improvement of precision for combined forecasts will be verylimited.更多还原