Video

What broker let’s you trade electricity?

Which platforms would you recommend for trading the US electricity market?

𝓟Ř𝔬𝓂𝔬𝐒ϻ 😥

Have there been any news on when the paper is going to be released?

𝔭𝔯𝔬𝔪𝔬𝔰𝔪 🎊

Wonderful, are there any r-code-examples for augmentation available?

Highly informative video. How do manage to incorporate covid-19 in ml model? What's the correct strategy to model covid-19 effects on revenue/sale forecast of household products?

Thanks for sharing the session.Would like to know is there any data analytics solution i.e. can be used to pull best bid/ask quotes in custom time interval of a day?, please let me know

Thank you. Very useful and John is master in this field. Hope to see more technical videos in future :)

Thank you Team. It will be awesome if Stephen Kolassa can take a live project in Forecasting done in Retail using R. All of us who wants to learn will pay and watch it. It will be awesome if this happens

This is just brilliant. Thanks a lot to Stephan for pointing out so many important facts about point forecasting that are - for so many reason I don't fully understand - mostly completely ignored by the forecasting community despite their importance.

Thank you for this video. It will be great if Stephan can make some more videos on Time series forecasting.

Dear speaker your conversation is disappeared could you appear it

Is there a way to get these slides?

Hi! Thanks for sharing. Are the slides uploaded somewhere?

P.S. Here's a good citation from (Goodwin, 2018, p. 54): "If you do want to perform FVA across products, then researchers recommend a measure called the average relative MAE (see the Davydenko and Fildes reference at the end of the chapter). This has not yet been implemented in commercial software." , see my previous comment for the references. *Source:* Goodwin, P. (2018). Profit from your forecasting software: A best practice guide for sales forecasters. Wiley and SAS Business Series. John Wiley & Sons, Inc., Hoboken, New Jersey.

Hi, guys! At 17:45 you talk about error metrics: 1) Importantly, it must be (Davydenko and Fildes, 2013), the year is 2013, not 2011 : Davydenko, A., & Fildes, R. (2013). Measuring forecasting accuracy: The case of judgmental adjustments to SKU-level demand forecasts. _International Journal of Forecasting_ , 29(3), 510-522 An adapted version of (Davydenko and Fildes, 2013) was published in this book: Davydenko, A., & Fildes, R. (2016). Forecast Error Measures: Critical Review and Practical Recommendations. In _Business Forecasting: Practical Problems and Solutions_ . John Wiley & Sons. 2) *FVA* you defined at 17:45 is exactly the *AvgRelMAE* from (Davydenko and Fildes, 2013). 3) *Bias* you defined at 17:45 is exactly the *AvgRelAME* proposed in the followng Ph.D. thesis on page 64: Davydenko, A. (2012). _Integration of judgmental and statistical approaches for demand forecasting: Models and methods_ (doctoral dissertation). Lancaster University, UK, doi.org/10.13140/RG.2.2.31788.62083 I really would appreciate it if you could cite (Davydenko, 2012, p.64) when you use the *AvgRelAME* for measuring and reporting bias. This metric was proposed in (Davydenko, 2012), I used my time, skills, and efforts to develop this metric for you. You are familiar with the source, so a proper citation is really required here. Thank you for your attention.

Additional responses from Sarah Darin to the questions asked during the webinar: Q: Would you recommend using events to help explain the big swings and combine that with domain knowledge with overrides? A: Yes, using a combination of modeling techniques with judgment is an excellent approach. Q: Do you recommend using ensemble forecasting models in dynamic regression models? A: Possibly, based on the evidence that ensemble approaches are effective. That said, I am more inclined to use something like Extreme Gradient Boosting as an ensemble approach as opposed to iterative Dynamic Regression. FYI, we are including Machine Learning methods in the next version of Forecast Pro, due out later this year. Q: How should we deal with the event factor? e.g. after the pandemic more and more people will choose online shopping, change their purchasing behaviour. A: I think this is inherent in the "new normal" that we discussed. Instead of an event, this really should be dealt with as permanent shift in the level. Extrapolative methods, such as exponential smoothing, should automatically account for a new level like this. Q: Where can I get more orientation on how to work with Forecast Pro dynamic regression modelling, when the automatic dynamics does not work well? (Ljung box parameter is not respected, for example) A: We are happy to work with you - please reach out to me at sdarin@forecastpro.com. We do provide ACF/PACF/error plots to guide this decision, making so that you can specify custom dynamics that have an appropriate Ljung Box test. A: Can you do forecasting in a "group" level e.g. a lipstick family where x% = Red, y% = brown, ... so 1 number at group level is then split per % to the lower levels. This would be done for forecasts 3+ months out. Within 3 months, we will go automatically to the sku's per color that could be maintained in detail. Do you have this functionality? A: You can forecast at a group level with a topdown modifier at any group level. We strongly encourage users to use approaches like this since very often it increases forecast accuracy. However, we do not let you specify a different model starting a given point in the forecast horizon - you need to choose sku level or group level. To achieve the forecast you describe, I recommend using the sku level forecast first, exporting those forecasts, reading those forecasts back into the project as external forecasts and then using the group level forecasts in the project. In the override grid, you can easily specify that the "external" sku forecasts be used for the first 3 months. Please take a look at www.forecastpro.com/resources/webinars/ where, if you scroll down a bit, you will find a webinar called "Effective Strategies for Forecasting a Product Hierarchy". Also feel free to reach out with questions sdarin@forecastpro.com. Q: Any insights on toll road traffic demand? Differences in short and long term? How to address possible permanent impacts? A: I certainly don't have a lot of domain knowledge, but I certainly think that some of the Google Mobility data metrics (hours spent at work, home etc.) would be helpful here. In terms of a permanent shift, a level shift is possible in dynamic regression, but I would be inclined to try and account for the drivers of the shift and then explore long run scenarios for those shift drivers (again things like google mobility, metrics that track working at home etc.) to generate a longer run forecast.

Very good overview of forecasting and demand planning. Thanks for sharing.