0
$\begingroup$

Mothly chart for 10 years

I am trying to forecast commodity price for next year. I have collected and plotted monthly average prices from last 10 years.Plot has been attached. I used Holt's-Winter method on prices till 2014 and forecasted prices for 2014-15. I did the same thing using ARIMA and TBATS method in R.

But none of the expected output comes close to actual output. I don't expect accurate forecast but I am looking for guiding forecasts.

     Actual     Winters                     Tbatsv                      ARIMA               
14-Mar  679 668 634 702 616 720     667 636 698 620 714     671 638 705 621 722
14-Apr  654 663 612 715 585 742     639 589 690 562 716     670 624 716 599 741
14-May  657 659 591 726 556 762     646 583 708 551 740     668 612 724 583 754
14-Jun  653 654 571 736 528 780     651 581 722 544 759     667 603 731 569 764
14-Jul  685 649 551 747 500 799     656 580 733 539 774     665 595 736 557 773
14-Aug  687 644 531 758 471 817     661 580 743 536 786     664 588 740 547 781
14-Sep  689 640 511 768 443 837     666 580 751 535 796     662 581 744 538 787
14-Oct  691 635 490 780 413 856     670 581 758 534 805     661 575 747 530 793
14-Nov  695 630 469 791 384 876     673 582 764 534 813     660 570 750 522 798
14-Dec  731 625 447 803 353 897     677 583 770 534 819     659 565 752 515 802
15-Jan  751 620 426 815 322 919     680 585 775 534 825     657 560 755 508 806
15-Feb  705 616 403 828 291 941     683 586 779 535 830     656 556 757 502 810

Is this the best forecast one can get? Is there any other method which suits for this kind of forecasting? Also if I want to forecast for 2015-16, how can I improve my model smoothing out 2014-15 forecasting errors?

$\endgroup$
  • 1
    $\begingroup$ Prices are notoriously hard to predict. Also forecasting 12 periods ahead usually increases error considerably. $\endgroup$ – mpiktas May 8 '15 at 10:46
  • $\begingroup$ Why there are 4 columns of data for each forecasting method? $\endgroup$ – mpiktas May 8 '15 at 10:47
  • 2
    $\begingroup$ If you are interested in the most accurate forecasts you can get, look at futures prices for these commodities. If you can beat that, you can become a millionaire by trading futures. (That means, there is very little chance you can actually beat that; there are too many large investment banks trying to do that.) Meanwhile, univariate time series methods like exponential smoothing and ARIMA models are not likely to do much better than a naive no-change forecast. That's the sad (?) reality, at least as far as I know. $\endgroup$ – Richard Hardy May 8 '15 at 10:48
  • $\begingroup$ @mpiktas when you have varying trend and seasonality in time series,are these the best model one can use? Is there any way I can smoothen out error using difference between actual and expected output? 4 columns - low 80% ,high 80%,low 95%,high 95% why attached picture is not visible?did I miss anything? $\endgroup$ – user2122922 May 8 '15 at 10:50
  • $\begingroup$ @RichardHardy I understand your point but with the dataset I have,which is the best suited model? $\endgroup$ – user2122922 May 9 '15 at 12:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.