I am currently working on a project in which I am trying to create a better forecast using extraneous variables, in particular a dummy variable representing whether a hurricane makes landfall to the US in a given month.

For example, let's say that I am given a monthly forecast (from a software package that uses a univariate approach taking into account 36 months of data and the item's seasonality and trends). For whatever reason, on this item the impact of a hurricane in a particular month is not picked up by the software's forecast.

With that being said, I would like to figure out the impact that a hurricane has on the demand for this item, and then add that result to the forecast. To do so, I would run a regression on the actual demand and lags of a hurricane dummy (lag 3 is the last lag to be statistically significant in this example) and find the impact at each lag. Then add these effects to the software's forecast as necessary when a hurricane occurs on a given month.

My question is, would this be valid way to attack this problem? If not, what are some other options for doing so? Any help is greatly appreciated!

Side note: from a practical perspective, from past business knowledge, it is known that a hurricane doesn't have impact on current months demand. Only future months. This would mean that for the purposes of my project, I don't need to predict whether a hurricane will occur in a month or not, just what impact this hurricane would have on future months after it has occurred.

Additionally, I know that there are a lot of other factors at play here (category of hurricane, where it makes landfall, etc.), but for now I just want to use the dummy variable mentioned above.

Thanks again!

  • $\begingroup$ Not sure if I believe the comment: "Side note: from a practical perspective, from past business knowledge, it is known that a hurricane doesn't have an impacted on current months demand. Only future months" as many people affected by hurricanes (now much more frequent) have learned from not having an ample supply of things like toilet paper, water,..., on hand. This change in inventory policy has likely created an artificial rise in demand (at least one time) for many products that were in short supply during a hurricane or other natural catastrophe events. $\endgroup$ – AJKOER Nov 30 '20 at 19:09
  • $\begingroup$ @AJKOER There is no consensus that hurricanes are becoming more frequent. There is some consensus that they are becoming stronger. Some say they are stronger but less frequent. This latter makes sense when you consider that a stronger storm covers a much larger area. That is, common sense dictates that fewer big marbles fit in the same jar as a larger number of small marbles. $\endgroup$ – Carl Dec 3 '20 at 7:45
  • $\begingroup$ @Carl: This hurricane season alone the number of named storms consumed the entire generated list of names. In fact, the contingency in such a case is to use the Greek alphabet (like Eta,...) and a good # of those. To quote a source: "The 2020 Atlantic hurricane season has truly been one for the record books. With 30 named storms, it has set the record for the most named storms in a single season." at scientificamerican.com/article/…) $\endgroup$ – AJKOER Dec 3 '20 at 11:56
  • 1
    $\begingroup$ @AJKOER Weather is not climate. I am well aware of the unusual weather this year. Climate is a long term average. I also recall a bad article that showed that for a dozen years there was no increase in temperature, that is also weather, not climate. Climate changes constantly and has for billions of years, not one year and not dozens. Over the last several hundred million years there have been periods of time with much higher CO2 and temperature. We are now in a period (long term) of increasing temperature and about half-way between the historical extremes of CO2 in the atmosphere. $\endgroup$ – Carl Dec 4 '20 at 4:02
  • $\begingroup$ So just use the landfall as the start of a clock, and see what happens thereafter. $\endgroup$ – Carl Dec 23 '20 at 23:37

Can you please provide a small example of you data? It is not clear how your data looks. what is item in your data? what are "lags of a hurricane dummy"? how the 3 lags are built in the data? do you mean something like: lag_1, lag_2,lag_3 which represents 3 last month, and 1 or 0 if hurricane was forecasted?

please provide an example so I can take better look.

  • $\begingroup$ Hurricanes have no reliable distribution so I am not sure how you would build them into time series. When they come, where, and how powerful they are would all matter. It is not, there is a hurricane, there is not. $\endgroup$ – user54285 Nov 30 '20 at 18:31
  • $\begingroup$ The data is historical data consisting of the following: the forecast for the demand of an item at a given month, the actual demand for that item in the given month, hurricane_dummy being whether a hurricane occurred in that month (either one or zero - even if multiple hurricanes occurred in a month, the value would remain 1). lag_1 being a dummy variable of whether a hurricane occurred in the prior month (if a hurricane occurred in January there would be a 1 for this variable in February) , etc. for lag_2 and lag_3. Does that help Arkady? $\endgroup$ – tfr950 Nov 30 '20 at 19:12

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.