Contact Us

How to Factor in Weather Impacts in Demand Forecasting

Jul 28, 2014 7 min

We have all been there. The first heat wave of the year is due to hit our towns and cities over the weekend, and you are hit by the urge to grab the charcoal and firelighters from the garage and buy some steaks for a proper barbecue. It’s a lovely prospect for most of us – day dreaming about good company, excellent food and a glass of wine in the garden. – For supply chain planners its less of an opportunity to daydream and more of a wakeup call to do some hard thinking about how to respond. Weather can have a drastic impact on consumer behaviour, so these responses can either have a transformative effect on the month’s bottom line or throw money down the drain.

Being able to take weather into account in supply chain planning is important throughout the product range. However, for most products it’s reasonably straightforward. Let’s take the classic example for that spring heat wave I mentioned; ice cream sales. Demand for ice cream can easily go up by 50% on a hot day and drop 20% if the weather is bad. What we have here is an asymmetric risk. If we haven’t filled up the freezers and the heat wave catches us by surprise it’s lost sales. However, if we do stock up but the forecast is wrong and the expected jump in demand fails to materialise, stores will have higher ice cream stock levels for a while but they will eventually sell out and the impact of lost revenue is small.

With fresh products, the game changes because of spoilage. As we discussed in our previous articles about managing fresh produce and short-shelf-life items, the risk of spoilage makes it crucial that you ensure your demand forecasting and replenishment are as accurate as possible even down to the level of forecasts for individual days. And while with ice cream we may have just pushed up stock levels a bit by filling up the freezers, with fresh meat a miscalculation might just as well mean filling up the rubbish bins with spoilt products on the following Monday. So having accurate forecasts for fresh products that take the impact of weather into consideration has a substantial business case. This article takes a look at the impacts weather can have on demand and how to take different scenarios into account.

Weather Dynamics

At the first glance the dynamics of how weather affects demand might appear complex. There are numerous components like temperature and sunshine, as well as factors like day-of-the-week, time-of-the-year, geography, type of product … However, it does not have to be overly complex. To simplify things; weather has two distinct kinds of effects on consumer behaviour.

  • People are more likely to go out more on a beautiful day, increasing the overall traffic and sales at the store level.
  • Consumer buying patterns being affected by weather, different products and product groups have different responses to weather changes.

Let’s discuss these two effects briefly. The impact of weather on overall traffic is most obvious during the summer with winter weather generally having far less impact (even in freezing Finland). A beautiful summer day can easily boost shop traffic by 5-10% compared to the seasonal average with weather factored out, while a cold rainy day has the opposite effect and reduces it. However the degree to which demand responds to the weather can differ considerably according to location and also market type. For example larger stores tend to feel the impact of a warm weekend more than smaller ones as people go to stock up for the weekend’s big barbecue party.

There’s also an enormous variation in the response to weather across different products and product groups. However, some factors remain more or less universal. The impact of good or bad weather is often asymmetric, as in the ice cream example above, because the potential sales increase as a result of good weather is stronger than the negative impact of bad weather. An example of weather impact is illustrated below. While there is quite a lot of noise in data, distinct significant impacts can be observed for the two largest heat waves.

Figure 1. Example of weather impact. Sales of all fresh products, ice creams and beverages (hence including most weather hot-weather sensitive products) for one store in June-July

However, both of the effects of the weather mentioned above have a zero-sum counter-part. Firstly, while people are more likely to go out to do their shopping on a beautiful day, conversely that means they will have, to some extent stocked up for other (perhaps less beautiful) days. So it’s apparent that the weather in part only shifts demand rather than just decreasing or increasing it. Secondly when sales of products that respond positively to hot-weather go up, people tend to spend slightly less on other products.

As well as the points made above, there are a number of subtle variations to the dynamics of how weather affects consumer behaviour. For instance responses to the weather may differ depending on the day of the week; prolonged good weather often desensitises consumers – the more beautiful days there are, the less people feel the need to make an event of them when they appear. However, the impact of such factors is of lesser significance. For instance, when separating out normal weekday variation from demand, the sales response to weather is typically reasonably uniform throughout the week.

So let’s next look at simple steps that you can take in order to start improving your forecast quality by taking the weather into account.

How Do I Model Weather Impacts in My Demand Forecasting?

Bearing in mind the various ways we’ve noted that weather can affect demand, three things need to be done to incorporate the weather into your forecasts.

  1. Segment your store locations based on responsiveness to weather
  2. Segment your low-level product groups based on responsiveness to weather
  3. Add weather modelling to your forecasts

1. Segment your store locations. Be pragmatic; focus at first on summer weather alone. A number of different factors need to be taken into account. Firstly, geographic segmentation is needed simply to account for regional weather differences. Secondly, store format has an impact (see above). Thirdly, there can be store-specific characteristics such as some locations being more sensitive to weather e.g. being in a tourist destination, an area with a high proportion of holiday homes, or being near a beach.

When your summer weather forecasting is working well you could create responses for other scenarios, such as extreme weather events. When there are high winds, heavy rain, flooding or unseasonably heavy snow forecast you may find, for instance, that it boosts sales ahead of the event as people stock up (not so much on ice creams as on canned goods and staples) and then suppresses sales as people put off shopping trips until trees are cleared, water has subsided or the roads have been gritted. Cross referencing your historic data against recent unusual weather events offers a useful benchmark.

2. Segment your (sub-) product groups. Statistical testing offers a good starting point. Providing you have sufficient sales and weather data it is possible to identify a unique weather response for a product at various hierarchy levels. However, if this approach is disproportionate in scale to the available data or potential benefit, just plain common sense and studying historic sales patterns combined with weather observations can take you a long way.

3. Adding weather modelling to your forecasts. What factors should you consider when forecasting based on weather? The temperature is an obvious choice and has the biggest impact but to take it one step further another variable is the ‘beautifulness of a day’, which can be measured e.g. with sunshine hours, cloud coverage or rainfall. And to do actual forecasting it’s not enough to consider just the absolute metric. Even more important is the difference to the seasonal average/normal – is it a nice or a horrible day for the time of year? By doing this, sales data can rather straightforwardly be de-weatherised and clean baseline sales be produced. Then the final step is to model back the upcoming weather forecast to the baseline sales forecast.

However, before going into advanced quantitative methodology, we should take one step back. As we’ve said, when discussing product-group segmentation, just studying past observations of sales against weather can provide valuable insights and be used to adjust forecasts. This becomes especially relevant for special events or holidays such as the UK’s late May or August bank holidays when sales of things to throw on the grill generally rocket, and where the exact calendar timing can also change from one year to another. In order to carry out your forecasting effectively, the weather information needs to be inputted to the supply chain planning system. In RELEX, one can freely pick and mix weather data and combine it with other data as illustrated in the figure below.

Figure 2. Example of reviewing product-group level forecast for different stores for Midsummer Eve

So far we have only discussed the demand forecasting aspect of taking weather into account. When planning replenishment, you also need to factor in lead times, the forecast horizon, and other constraints. ‘What is our capability to react to forecasted heat wave for the weekend?’ ‘Do we have enough balance and picking capacity at the DC?’ ‘How much available shelf space do the stores have?’ ‘How much uncertainty is there regarding the weather forecast?’

One of the most powerful features RELEX systems offer is the ability to simulate your upcoming deliveries, orders, stock levels, and spoilage throughout the supply chain based on unified forecasts. Let’s go back to the scenario of a heat wave predicted to hit the city the next weekend. Let us assume you have modified your forecasts to match the heat wave but you are not sure if you want to take the risk in case the weather suddenly reverts to the seasonal norm. With RELEX you can bottom-up simulate how your supply chain would behave if the original sales forecast did not hold. Hence, it is possible to directly see what would e.g. be the financial loss due to potential spoilage should the weather change.

All in all, when weather observations and data are fully integrated into the planning system, it enables the supply chain planners to consider and analyse weather related items effectively. So, right now, summer is still a few months away, so better get started and plan how to respond whether it brings you sunshine or rain.

Written by

Tommi Ylinen

Tommi Ylinen

Chief Product Officer