```Date: Tue, 23 Aug 2011 19:57:23 -0400 Reply-To: Francois van der Walt Sender: "SAS(r) Discussion" From: Francois van der Walt Subject: Re: Time series data analysis - help required. Comments: To: Shashank Sawant Dear Shashank, I do not have any real comment on all the previous advice, that might be valid. I do see myself as a practicing statistician that deals with very similar problems on a daily basis for the various clients that GJI serve. If you want a solution rather than a complex difficult to explain process I suggest you look at Process control charts, also known as XmR or "Mean and Range" or Shewhart charts. SAS do have a module (Proc Shewhart) that can give you results that you might want. (I do not have it licenced so have to resort to my own Excel developed Charts). I did for an exercise worked through your data and came up with the following: 1) The data is unpredictable. The average premium per month is: \$6,941. It can vary from -\$4,692 (Lower Natural Process Limit(LNPL) to \$18,574 (Upper Natural Process Limit(UNPL)) 2) The following months and values can be seen as outliers (outside natural process limits): Nov-08 \$52,641 Nov-09 \$21,321 Dec-09 \$36,468 Nov-10 \$43,563 Dec-10 \$20,196 Removing them from the calculations the premiums is still unpredictable. Without these months the average Premium is expected to be: \$3,270 with the LNPL=-\$3,032 and the UNPL=\$9,572. Using the excluded outliers it can be shown that the average Premium for Nov and Dec is: \$34,838. The growth (from my calculations) for these months (November and December) is 16% per year. 3) In my analysis I worked through the above another 2 times (removing outliers and revise the predicted value with the Natural Process Limits) and did not find any evidence that there is growth of premiums for the other months. The proposed technique I used is highly graphical and sadly I cannot post the graphs on this forum. If you would like to see it please let me know and I will send it to your email address with the graphs. I found the above procedure easy to communicate to business owners and other stakeholders. I found the general timeseries techniques hard to communicate. If you want to read more about this technique I can refer you to Donald Wheeler's book: "Understanding variation - The key to manage Chaos" or his other book: "Making Sense of Data". Kind Regards Francois van der Walt (Brisbane) Senior Business Analyst www.gji.com.au On Sun, 21 Aug 2011 21:03:34 +0530, Shashank Sawant wrote: >> Thank you All for your reply . Below is the attempt that I made to >> proceed: >> >> I ran proc arima to check for autocorrelation and stationarity >> >> Autocorrelation Check for White Noise >> To Chi- Pr > >> Lag Square DF ChiSq >> ------------------Autocorrelations----------------- >> 6 6.00 6 0.4230 0.166 -0.110 -0.191 -0.196 >> -0.038 -0.089 >> >> Augmented Dickey-Fuller Unit Root Tests >> Type Lags Rho Pr < Rho Tau Pr < >> Tau F Pr > F >> Zero Mean 0 -25.7891 <.0001 -4.29 <.0001 >> 1 -23.9950 0.0002 -3.40 0.0011 >> Single Mean 0 -34.9700 0.0003 -5.33 0.0002 >> 14.19 0.0010 >> 1 -45.7168 0.0003 -4.59 0.0006 >> 10.56 0.0010 >> Trend 0 -35.2933 0.0002 -5.29 0.0005 >> 14.02 0.0010 >> 1 -46.7970 <.0001 -4.59 0.0036 >> 10.53 0.0010 >> The results show that there is no auto correlation and the series is >> stationary.I think this suggest that I don't need to build any ARIMA model. >> So, As per your suggestion, I ran below procedure,as after every 6 months >> there is significant increase in the premium.Also, it is expected That in >> Nov2011 the premium figure would increase by 8% from Nov2010 figure.How do I >> incorporate this information to >> forecast future values? >> * >> >> proc >> **forecast* data = new interval = month method =addwinters seasons = *6*lead = >> *6* >> >> out = pred; >> >> var >> sales; >> >> id >> year;* >> >> run >> *; >> The output I got is >> year _TYPE_ _LEAD_ prem 11-Aug FORECAST 1 1749.157 11-Sep FORECAST 2 >> 1019.377 11-Oct FORECAST 3 2428.48 11-Nov FORECAST 4 17654.16 11-Dec >> FORECAST 5 12600.83 12-Jan FORECAST 6 2289.852 >> Now the question is how do I validate the results? >> >> Any views, on this would be of great help. >> >> Thanks in advance. >> >> Regards, >> Shashank >> >> >> On Sun, Aug 21, 2011 at 3:09 PM, Tansu TOSUN wrote: >> >>> Hi Shashank , in addition to Wensui's and Murphy's comments, you could >>> think to add external factors that cause this situation (your data seems >>> so >>> volatile). Seasonal ARIMA + External factors could be a trial. Then you >>> could try model combination with forecast procedure. >>> >>> >>> >>> On Sun, Aug 21, 2011 at 1:10 AM, Wensui Liu wrote: >>> >>> > hey Shashank >>> > >>> > It is a little unrealistic to use arima model forecast 5 months out. >>> > this is a common mistake that usually a newbie would make. >>> > >>> > generally, arima is a short-memory model and best used to predict a >>> > couple periods ahead. >>> > >>> > for your case, your better bet might be forecast procedure. however, >>> > an interesting observation is that the forecast result is sensitive to >>> > the default setting of this procedure :-). >>> > >>> > >>> > On 8/20/11, Shashank Sawant wrote: >>> > > Hi All, >>> > > >>> > > I have a time series data and need to forecast premium from >>> Aug11-Dec11. >>> > I >>> > > have gone through PROC ARIMA, FORECAST reading materials, but unable >>> to >>> > > proceed. >>> > > Could you please help me on how to proceed further/ >>> > > >>> > > Below is the data for your reference. >>> > > >>> > > Thanks in advance for your help. >>> > > >>> > > data new; >>> > > input year:monyy. premium @@; >>> > > format year monyy.; >>> > > datalines; >>> > > Jan08 9909 >>> > > Feb08 8548 >>> > > Mar08 191 >>> > > Apr08 858 >>> > > May08 143 >>> > > Jun08 6244 >>> > > Jul08 5007 >>> > > Aug08 1721 >>> > > Sep08 644 >>> > > Oct08 386 >>> > > Nov08 52641 >>> > > Dec08 5463 >>> > > Jan09 215 >>> > > Feb09 658 >>> > > Mar09 510 >>> > > Apr09 4832 >>> > > May09 485 >>> > > Jun09 2382 >>> > > Jul09 265 >>> > > Aug09 1006 >>> > > Sep09 6302 >>> > > Oct09 2178 >>> > > Nov09 21321 >>> > > Dec09 36468 >>> > > Jan10 1833 >>> > > Feb10 4999 >>> > > Mar10 2096 >>> > > Apr10 1749 >>> > > May10 3215 >>> > > Jun10 9953 >>> > > Jul10 -108 >>> > > Aug10 1342 >>> > > Sep10 1830 >>> > > Oct10 9677 >>> > > Nov10 43563 >>> > > Dec10 20196 >>> > > Jan11 9445 >>> > > Feb11 -618 >>> > > Mar11 788 >>> > > Apr11 2428 >>> > > May11 3431 >>> > > Jun11 13550 >>> > > Jul11 712 >>> > > ; >>> > > run; >>> > > Regards, >>> > > Shashank >>> > > >>> > >>> > >>> > -- >>> > ============================== >>> > WenSui Liu >>> > Credit Risk Manager, 53 Bancorp >>> > wensui.liu@53.com >>> > 513-295-4370 >>> > ============================== >>> > >>> >>> >>> >>> -- >>> -------------------------------------------------------------- >>> Tansu TOSUN >>> >> >> ```

Back to: Top of message | Previous page | Main SAS-L page