第四章時間序列分析法ppt課件

1、市場預(yù)測與管理決策Forecasting Market and Decision Making馮燕1862 7016 225: 6326 48182本章構(gòu)造 時間序列分析法的特點(diǎn)與步驟1.簡易平均法2.挪動平均法3.指數(shù)平滑法4.趨勢延伸法5.4.1 定義、特點(diǎn)與步驟什么是時間序列分析法？時間序列分析法的特點(diǎn)：根據(jù)過去變化趨勢，預(yù)測未來開展時間序列數(shù)據(jù)變動存在著規(guī)律性和不規(guī)律性撇開市場開展的因果關(guān)系時間序列市場預(yù)測法的步驟什么是時間序列分析法？時間序列：市場景象的統(tǒng)計(jì)目的數(shù)值，按時間先后順序陳列而成的數(shù)列。時間序列分析法：分析和研討，建立預(yù)測模型，預(yù)測編制時間序列要

2、做到：總體范圍一致；代表的時間單位長短一致；統(tǒng)計(jì)數(shù)值的計(jì)算方法和計(jì)量單位一致。 Time Series Analysisuse past data to forecast futurepast data should be availableuseful for short termuseful for stable datauseful for forecasting for large number of itemscan not predict turning points (lag effect)時間序列分析法的特點(diǎn)一、前提是假定事物的過去會延續(xù)到未來。 未來開展過去歷史的簡單反復(fù) 短

3、期市場預(yù)測 中長期市場預(yù)測 時間序列分析法的特點(diǎn)二、時間序列數(shù)據(jù)變動存在著規(guī)律性與不規(guī)律性 長期趨勢變動T 季節(jié)變動S 循環(huán)變動C 不規(guī)那么變動I 乘法模型：時間序列分析法的特點(diǎn)三、撇開因果關(guān)系 將一切的影響要素歸結(jié)到時間這一要素上，只成認(rèn)一切影響要素的綜協(xié)作用 時間序列分析法的步驟搜集、整理時間序列，繪制圖形對時間序列進(jìn)展分析選擇預(yù)測方法，建立預(yù)測模型測算預(yù)測誤差誤差度量絕對相對平均平均絕對均方差百分平均百分平均絕對百分4.2 簡易平均法簡單算術(shù)平均法加權(quán)平均法幾何平均法Example:Forecasting at FastchipsFastchips is a leading produc

4、er of microprocessors.Six months ago, it launched the sales of its latest microprocessor.Month-by-month sales (in thousands) over the initial six months have been172524263028Question: What is the forecast for next months sales?The Last-Value Forecasting MethodThe last-value forecasting method ignore

5、s all data points in a time series except the last one.Forecast = Last valueFastchips: Month-by-month sales (in thousands) over the initial six months:172524263028Forecast = 28Reasonable when conditions tend to change so quickly that sales before the last months are not a reliable of future sales. 簡

6、單算術(shù)平均法將察看期內(nèi)預(yù)測目的時間序列值加總平均，求得算術(shù)平均數(shù)，作為下期預(yù)測值。時間序列數(shù)據(jù)方差越小，簡單平均數(shù)作為預(yù)測值的代表性越好。缺陷：一切察看值不論新舊在預(yù)測中一概同等對待，這不符合市場開展的實(shí)踐情況。 The Averaging Forecasting MethodThe averaging forecasting method uses all the data points in the time series and simply averages these points.Forecast = Average of all data to dateFastchips: Mo

7、nth-by-month sales (in thousands) over the initial six months:172524263028Forecast = (17+25+24+26+30+28) / 6 = 25Reasonable when conditions tend to remain so stable that even the earliest sales reliable indicator.加權(quán)平均法根據(jù)察看值重要性不同對影響大的近期察看值給予較大的權(quán)數(shù)，對影響小的遠(yuǎn)期察看值那么給予較小的權(quán)數(shù)，分別給予相應(yīng)的權(quán)數(shù)，再計(jì)算加權(quán)平均數(shù)作為建立預(yù)測模型的方法。權(quán)數(shù)確實(shí)

8、定：距預(yù)測期的遠(yuǎn)近，動搖幅度大小缺陷：對于趨勢變動明顯的時間序列，無論怎樣的加大權(quán)數(shù)， 也跟不上實(shí)踐值的變動，它小于后期的實(shí)踐察看值，更不能作為預(yù)測值。 幾何平均法 計(jì)算出一定時期內(nèi)預(yù)測目的時間序列的開展速度或逐期增長率，然后在此根底上進(jìn)展預(yù)測。適用于趨勢變動規(guī)律表現(xiàn)為開展速度一樣的時間序列。4.3 挪動平均法挪動平均法的概念和特點(diǎn)一次挪動平均法加權(quán)挪動平均法挪動平均法的概念和特點(diǎn)對時間序列察看值由遠(yuǎn)到近按一定跨越期計(jì)算平均值的預(yù)測方法。適宜于既有趨勢變動，又有動搖的時間序列。一次挪動平均法由延續(xù)挪動構(gòu)成的各組數(shù)據(jù)，用算術(shù)平均法計(jì)算各組數(shù)據(jù)的挪動平均值。缺陷：只能向未來預(yù)測一期對于有明顯趨勢變

9、動的市場景象時間序列不適宜。由于一次挪動平均值大大滯后于實(shí)踐察看值。The Moving-Average Forecasting MethodThe moving-average forecasting method averages the data for only the most recent time periods.n = Number of recent periods to consider as relevant for forecastingForecast = Average of last n valuesFastchips: Month-by-month sales

10、(in thousands) over the initial six months:172524263028Forecast (n=3) = (26+30+28) / 3 = 28 17252426302828Forecast (n=3) = (30+28+28) / 3 = 29Reasonable when conditions tend to change occasionally but not extremely rapidly.加權(quán)挪動平均法對市場景象察看值按距預(yù)測期的遠(yuǎn)近給予不同的權(quán)數(shù)，并按其加權(quán)計(jì)算出挪動平均值。The Weighted Moving-Average Fore

11、casting MethodThe weighted moving-average forecasting method averages the data for only the most recent time periods with weighted.n = Number of recent periods to consider as relevant for forecastingForecast = Weighted Average of last n valuesFastchips: Month-by-month sales (in thousands) over the i

12、nitial six months:172524263028Forecast (n=3, w1=0.2, w2=0.3, w3=0.5) = (26*0.2+30*0.3+28*0.5) = 28.217252426302828.2Forecast (n=3 , w1=0.2, w2=0.3, w3=0.5) = (30*0.2+28*0.3+28.2*0.5) = 28.54.4 指數(shù)平滑法指數(shù)平滑法的概念和特點(diǎn)一次指數(shù)平滑法指數(shù)平滑法的概念和特點(diǎn)特殊的加權(quán)挪動平均法。特點(diǎn)：離預(yù)測期最近的察看值給予最大的權(quán)數(shù)察看值對預(yù)測值的影響由遠(yuǎn)及近按等比數(shù)列減小，其首項(xiàng)是 ，公比為 。預(yù)測值可以經(jīng)過調(diào)整

13、 的大小來調(diào)理近期察看值和遠(yuǎn)期察看值對預(yù)測值的不同影響程度。 The Exponential Smoothing Forecasting MethodThe exponential smoothing forecasting method places the greatest weight on the last value in the time series and then progressively smaller weights on the older values.Forecast = a (Last value) + (1 a) (Last forecast)a is the

14、 smoothing constant between 0 and 1.The choice of the value of the smoothing constant a has a substantial effect on the forecast.A small value (say, 0.1) if conditions are relatively stable.A larger value (say, 0.5) if significant changes occur frequently.ESFt = Ft-1 + (Dt-1 - Ft-1)Example: = 0.2, L

15、et F1 = 1000F2 = 1000 + 0 = 1000F3 = 1000 + 40 = 1040F4 = 1040 + 172 = 1212F5 = 1212 - 82.4 = 1129.6Periods12345678Actual Demand10001200190080011201350?FtF1F2F3F4F5F6F7F8ESFt = Ft-1 + (Dt-1 - Ft-1)Example: = 0.2, Let F1 = 1000F6 = 1129.6 - 1.92 = 1127.68F7 = 1127.68 -44.46 = 1172.14Similar to weight

16、ed moving average exponentially decreasing weights for all previousperiodsPeriods12345678Actual Demand10001200190080011201350?FtF1F2F3F4F5F6F7F8Forecast ErrorMean Absolute Error (MAE)Root Mean Square Error(RMSE)二次挪動平均法二次挪動平均法是對時間序列一次挪動平均值再進(jìn)展第二次挪動平均，利用第一次挪動平均值和二次挪動平均值構(gòu)成時間序列的最后一個數(shù)據(jù)為根據(jù)建立線性模型進(jìn)展預(yù)測。二次挪動平均

17、法二次挪動平均法的特點(diǎn)：二次挪動平均法與一次挪動平均法相比，其優(yōu)點(diǎn)是大大減少了滯后偏向，使預(yù)測準(zhǔn)確性提高。二次挪動平均只適用于短期預(yù)測，并且時間序列數(shù)據(jù)呈現(xiàn)線性趨勢變化的預(yù)測。二次挪動平均法比一次挪動平均法適用面更廣，在實(shí)際中運(yùn)用較多。二次挪動平均法例7-4 由于歷史數(shù)據(jù)根本呈線性趨勢，且又有動搖，為靈敏反映其變動趨勢，挪動平均的跨越期宜短一些，設(shè)n=3兩次挪動的n應(yīng)取值一致 1.計(jì)算一次和二次挪動平均值一次挪動平均值：二次挪動平均值：2.計(jì)算各期a，b 值3.計(jì)算察看期內(nèi)估計(jì)值4.運(yùn)用預(yù)測模型計(jì)算預(yù)測值應(yīng)該留意的是，察看期內(nèi)各期估計(jì)值的a，b值不同，而在預(yù)測期各預(yù)測值的a，b值是一致的，即最后一個察看期的a，b值。上例中a