2
$\begingroup$

I have an exercise given to me during the recruitment process which asks to calculate the next year's January hourly forward prices for an asset given the historical hourly prices of that asset and next year's January average price (average of all hours in that period). The exercise asks to take account of the seasonality (modelling seasonality), hence there is a 9 year hourly price dataset given.

I have studied statistics and econometrics at school but have forgotten most of the techniques that could be used for price estimation.

Any idea how this exercise can be solved? I can provide a sample dataset on request.

$\endgroup$
  • $\begingroup$ Since this is a de facto homework question, it should be tagged as such. $\endgroup$ – rocinante May 23 '14 at 11:54
  • $\begingroup$ I don't believe that this is homework at all. I Think that a challenge was laid out by an individual (the recruiter) who was unaware of this website. I would treat the question as coming from them and in this light it is definitely not homework. $\endgroup$ – IrishStat May 23 '14 at 12:14
  • 2
    $\begingroup$ Sorry, but I disagree. This is a screening question for a job offer that means that the knowledge base of the candidate is being evaluated. If he does not know, it is dishonest to pretend that he does by just foisting off the question to other people, with zero effort on his part to attempt a solution. The whole point of the screening questions is to weed out people who don't know. Lots of professors don't know about this site either; that does not mean it is not unethical and a violation of student conduct if students post their assignment questions on SE without disclosing the context. $\endgroup$ – rocinante May 23 '14 at 13:13
  • 1
    $\begingroup$ I have seen the recruiting effort actually test not only the knowledge of the prospect BUT the resourcefulness of the prospect. It is one thing not to know and not to know where the solution might be AND not to know but also to know where to get ethical help to deal with the question. I believe in attempting to help people who ethically seek help and to provide guidance as they attempt to get smarter about things. I trust you are also of this mind and emulate @whuber as he eminently pursues this goal. $\endgroup$ – IrishStat May 23 '14 at 20:48
  • $\begingroup$ quant.stackexchange.com would be a better place to ask this question $\endgroup$ – lcrmorin Jan 1 '15 at 16:00
2
$\begingroup$

We have studied very similar problems mostly in the retail sector. One needs to recognize that the eventual hourly forecasts need to aggregate/reconcile to the monthly forecast. We have implemented a two-prong approach which uses monthly values to predict hourly values taking into account monthly patterns and hourly effects. These models also incorporate any level shifts or trends that may be found to be statistically significant. Additionally any auto-regressive structure is also incorporated ( previous prices). Care is also taken to identify anomalous data and adjusting for these effects to ensure a robust solution. Known events such as Holidays and other possible causes are also included/tested in a very sophisticated manner. The software package I have helped develop called AUTOBOX implements this solution. I would suggest that you review it ( there is a free 30 day trial ) and if you like the results either acquire it or rewrite it in any language/system that you wish. Finally one needs to validate that the final model has an error process that has constant variance and that the model's parameters are invariant over time.

In closing the recruiter is simply trying to find that employee that can solve a problem that they can't currently solve OR is as creative/ingenious as your are to be able to reach out to subject matter experts and acquire/find an answer. If I were them I would definitely hire you !

$\endgroup$
  • $\begingroup$ Thank you for your reply! I will definitely try the Autobox out. As for the exercise, as I stated previously, this task is not just to predict prices based on historical values, but there is also a monthly average constraint already given for the period that needs hourly data to be calculated. So I reckon one needs to use a Solver function for Excel to change the hourly prices in such a way that would solve the equation. Only question is which conditions have to be met for Solver to work properly? $\endgroup$ – Erik May 23 '14 at 12:31
  • 1
    $\begingroup$ @Erik If truly have forgotten the techniques, then the thing for your to do is to re-read your class/text notes. The solution suggested above does not apply to forecasting forward prices of stocks at all. $\endgroup$ – rocinante May 23 '14 at 13:19
  • $\begingroup$ It's not that I haven't made any attempts to solve it - I have, but I realize the result is not close to reality. By writing here, I was looking for some guidance which in my opinion is a normal practice. Networking is efficient after all. The fact that I can't solve it by myself at present does not mean I don't want to learn how to do it. $\endgroup$ – Erik May 23 '14 at 13:24
  • $\begingroup$ @rocinante I agree that stock price forecasting is difficult BUT for you to dismiss the general approach is a bit much in my opinion as there may be changes in volatility that can be identified and/or particular days of the week/month that might be important. Don't throw out the baby with the bathwater ! Take the structure that you can identify and move forward. It is better to light a candle than curse the darkness. $\endgroup$ – IrishStat May 23 '14 at 13:45
  • $\begingroup$ @Erik The problem here is three-fold. One, you misrepresented your efforts by claiming that you "forgot", when you now just admitted that you never knew how to do it in the first place. Second, SE rules are clear that attempts at solution should be disclosed, along with any data/code output. Thirdly, as a job screening question, it is unethical to crowd-source a solution. You are the one being hired for the job, not SE members. $\endgroup$ – rocinante May 23 '14 at 13:46

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.