AgPa #23: Trading on the Weather

Global weather-based trading strategies (2022)
Ming Dong, Andréanne Tremblay
Journal of Banking & Finance, Volume 143, 106558, URL/SSRN

People tend to be in a better mood when the sun is shining. I think most of us agree with that and there is even a biological explanation for it. Exposure to sunshine leads to more serotonin, a hormone associated with better mood and calmness.[1]More information for example here. This week’s AGNOSTIC Paper builds on this effect and examines if the weather is related to stock returns around the world. In the end, investors are just humans like everybody else. So they may be more willing to buy stocks on a positive sunny day.

Everything that follows is only my summary of the original paper. So unless indicated otherwise, all tables and charts belong to the authors of the paper and I am just quoting them. The authors deserve full credit for creating this material, so please always cite the original source.

Setup and Idea

The setup and main idea of the paper is very simple. If weather affects the mood of investors (or any other relevant variable), there could be a correlation between weather data and stock returns. The results of the literature on this issue have so far been mixed. However, the authors of this week’s paper did (to the best of my knowledge) the most comprehensive study on the topic. In particular, they use global data and not only information about sunshine but also four other weather variables.

Although some earlier studies doubt that the relation between weather and stock returns really exists, examining weather data has one important advantage. It is truly exogenous to all types of economic or financial activity. Firms can optimize their fundamentals to please investors but nobody can change the weather (at least until now).

Long story short, the authors test if weather data can serve as a profitable signal for real-world investors. Specifically, the suggest two different trading strategies based on country stock market indices. More on that below.

Data and Methodology

The heart of the analysis is of course the weather data. The authors get this data from the Integrated Surface Database (ISD) which provides hourly weather conditions from various weather stations around the world. Specifically, they look at the following five weather variables.

  • Sky Cover (SKC): a categorical variable for the cloudiness of the day. Possible values are clear sky, scattered cloud cover, broken cloud cover, and overcast sky.
  • Wind Speed (WIND): wind speed in miles per hour.
  • Temperature (TEMP): temperature in degrees Fahrenheit.
  • Perception (RAIN): a dummy variable indicating if there is any liquid precipitation.
  • Snow Cover (SNOW): snow cover in inches. Unsurprisingly, this variable is only relevant for some countries and some periods of the year.

For each country, the authors use data from the weather station that is closest to the countries’ main exchange (New York for the US, Frankfurt for Germany, etc.). To bring the data on a daily scale, they use averages of hourly data between 6:00AM and 4:00PM. This is roughly in-line with the opening hours of most exchanges and thus a reasonable assumption.

In the next step, they match the weather data with returns of 49 country stock market indices from Datastream (now Refinitv), a high quality data provider. So there shouldn’t be any issues with this data. Since the impact of weather likely also depends on the overall climate conditions, the authors additionally group the 49 countries into Temperature Regions.[2]Nobody is happy about a sunny day in Australia, it’s just normal. But if you have a sunny day during November in Germany, this is really cool. The table summarizes the sample and those temperature regions.

Excerpt of Table 2 of Dong and Tremblay (2022).

For their trading strategies, the authors first split the sample into a training- and out-of-sample validation-dataset. They use the period between 1973 and 1992 to run OLS regressions of daily stock returns on the five weather variables within each temperature region and calendar month. The separate regressions are necessary to control for region- and seasonal effects.[/fn]A sunny day in Winter is much nicer than the 17th sunny day in a row in summer.[/fn] After 1992, they add each passed year to the training-data such that the estimation get more precise over time.

In the next step, the authors use their regression equations to predict stock returns out-of-sample. Specifically, they use the hourly average of the weather variables between 5:00AM and 9:00AM (something like pre-market weather) to determine a return estimate for each index every day. This is important because when the market opens at 9:00AM, you have your return estimate and can trade on it. But it is of course also not fully consistent because the model is estimated on the average weather between 6:00AM and 4:00PM. Nonetheless, it’s a reasonable way to avoid look-ahead biases. The daily return estimates are then the signals for the following two hypothetical trading strategies.

Important Results and Takeaways

The global long-short weather strategy returned 15.2% p.a. between 1993 and 2012

For their first strategy, the authors go long the country index with the highest predicted return and short the one with the lowest (Hedged strategy). They rebalance this portfolio daily. This sounds like a lot of turnover (which is true), but given that the strategy involves stock indices, this should be relatively easy and cheap to implement via futures or ETFs. At this point, I should also mention that the regressions indeed suggest a positive relation between weather and stock returns. So essentially, the strategy goes long the index from the country with the best weather on the respective day. The following table summarizes the hypothetical performance between 1993 and 2012.

Excerpt of Table 4 of Dong and Tremblay (2022).

The long-short strategy returned 15.2% per year and generated a Sharpe ratio of 0.46. It also generated statistically significant alpha with respect to the Datastream World Equity Index. The magnitude of those returns are remarkable and the strategy easily outperformed most stock indices. The first three columns also show results for the temperature region subsamples. The results were best for the cold countries, however, this is probably also driven by the fact that some of this regions performed generally better.

Note however, that these results are before trading costs. The rebalancing ratios indicate the fraction of days where either the long- or short-side must be rebalanced. This number exceeds 100%, meaning that the long- and short-side are essentially changed every day. Despite this turnover, the authors argue that the annual return of 15.2% is so large that the strategy should remain profitable even after trading costs. They estimate daily costs of around 1 basis point as the strategy can be implemented via index futures or ETFs. Assuming 250 trading days and the rebalancing ratio of 150%, this leads to a performance loss of about 3.75% per year.[3]250 x 0.01% x 150% = 3.75% This yields an annual net return of about 11.5% which is indeed still very good.

The long-only version of the strategy returned 13.4% p.a.

For their second strategy, the authors simply skip the short leg (Long-Only strategy). They just go long the country index with the highest predicted return, i.e. buy the index from the country with the best weather on the respective day. This reduces the turnover and complexity (no need of shorting) of the strategy but potentially also sacrifices profits from the short leg. The following table again summarizes the results.

Excerpt of Table 6 of Dong and Tremblay (2022).

Annual returns are with 13.4% still very high but lower than for the long-short version. The strategy also earns lower risk-adjusted returns as shown by the smaller Sharpe ratio of 0.36. The alpha with respect to the Datastream World Equity Index is also lower but still statistically significant. However, the rebalancing ratio more than halved from 150% to 73% indicating much lower turnover. Using again 250 trading days and 1 basis point trading costs per day, this leads to a performance loss of about 1.83% per year (vs. 3.75% before). After trading costs, the two strategies are therefore quite comparable.

The following chart again summarizes the performance of the strategies and plots the cumulative returns (before costs) as well as the Datastream World Equity Index and the CRSP value-weighted index (US market). The important takeaway (in my opinion) is that although the strategies outperformed, the ride has been anything but smooth. Especially for the long-short strategy, there are some severe and long drawdowns. So trading on the weather was indeed profitable, but it was by no means easy to stick with those particular strategies.

Figure 1 of Dong and Tremblay (2022).

Conclusions and Further Ideas

I completely understand that some of you are skeptical about the results and the suggested mechanism between weather and investors’ mood. I mean, explaining stock returns only by the pre-market weather just seems too simple, right? What about the news and events of the day? What about the correlation between stock markets of different countries? These are all valid points and I certainly wouldn’t commit all my capital to such a simple weather-strategy.

However, I do believe that the weather is an interesting data source that is worth exploring. Especially the fact that it’s purely exogenous and available in real-time makes it an interesting addition to other sentiment variables like news or social media. I also believe that the behavioral mechanism is at least to some extent plausible. All else equal, most of us are probably happier when the sun is shining.[4]I rarely hear someone saying, “Oh, this rainy day was so beautiful.” I see no reason why this shouldn’t apply to people who are responsible for investments. Admittedly, however, it remains very hard or maybe even impossible to properly isolate such a mechanism in the data. In addition to that, I also have some doubts about the findings because the authors don’t really present convincing robustness tests.[5]Unfortunately, there is not much except for “the results are too strong to be random”.

Despite those problems, the results are still interesting and there are many ways to refine the strategies in practice. We could, for example, add weather data to other sentiment variables like news. Another way is to address the simplifying assumptions of the paper. For example, the authors always use the weather closest to the major exchange. This is completely reasonable but not necessarily the whole picture. In the US, for instance, the major exchange is of course New York but there are many asset managers (who ultimately trade and move prices) in Boston, Chicago, or Los Angeles. These are minor details but the overall idea remains unchanged. If the weather is good, stock returns tend to be good as well and this is something we can try to exploit.

This content is for educational and informational purposes only and no substitute for professional or financial advice. The use of any information on this website is solely on your own risk and I do not take responsibility or liability for any damages that may occur. The views expressed on this website are solely my own and do not necessarily reflect the views of any organisation I am associated with. Income- or benefit-generating links are marked with a star (*). Please also read the Disclaimer.