YouGov share price, hit movies and election results

If, like me, you took your eye off the ball at the end of last week, you may have missed that YouGov ’s share price halved overnight due to a profit warning and is now at a 5 year low (it was my birthday, that’s my excuse). 

Superficially, you might expect an election year to be good for a famous global polling company, but as is widely reported , political polling makes up only a small proportion of YouGov's revenues. 

The real story relates to a topic I covered in my my newsletter a few months ago on data firm valuations. YouGov’s elevated valuation multiple was based on expectations for future growth - in particular for its data products. So when growth starts to falter, that doesn’t just hit the earnings for the year, but also dampens the expectation for growth going forward. Worse still, if as appears to be the case, data products are suffering more than customised research, that further undermines the overall growth story, and YouGov starts to look like a traditional research firm. As of today, at 20x earnings Yougov's P/E ratio is still high, and higher than Ipsos at 17x but the gap has closed significantly. 

This share price story potentially relates to a broader, very topical question of the accuracy of political polling and prediction science more generally. I am an avid listener to the funny and informative podcast The Rest Is Entertainment with Marina Hyde and Richard Osman. In the most recent episode, Marina opines that “the science of polling is creaking”. She is referring specifically to predicting the popularity of movie openings, but sees a read-across to political polling. 

On one level political polling has become more sophisticated and scientific than ever before with MRP seat-by-seat models becoming the methodology du jour, such as this example from Ipsos . My layman's understanding of MRP models is that pollsters analyse previous election results by seat and by demographic, then calculate the probability of variances to those results based on up-to-date survey data. So for instance, if 45-year-old women in Guildford (to pick a random example) voted Conservative in the last election, but the survey suggests that group's voting intention is shifting to Lib Dem, the model probability is updated accordingly. Multiply that up across all constituencies and demographics, and you have an MRP model.

This methodology requires sampling dozens of carefully targetted voters in each constituency, so overall requires a much bigger sample nationally than an ordinary poll - upwards of 15,000 respondents compared to under 2,000 for a standard poll.

So with all this survey data and statistical science available, why might polling be "creaking"? In the good old days pollsters were able to rely on quite stable demographic and media consumption models. Newspaper readership, for instance, was a good predictor of voter intention. If you knew how Sunday Times readers were likely to vote, and you knew what proportion of the population they made up, that allowed you to extrapolate your survey results with reasonable accuracy. And similarly, if ITV viewers liked a particular movie, you might predict a box-office hit.

Today of course, media consumption is much more fragmented across social media, streaming platforms and digital devices. Tracking these audiences, the messages they are exposed to, and matching that back to behaviour has become much harder for pollsters, politicians and movie executives. Who knows how Tik Tok users are planning to vote, or what movies they like and is that even a useful thing to measure?

The other factor is that with the rise of disruptive populist political movements, voters have simply become more unpredictable, and historic dividing lines no longer offer a reliable guide to future intention. So when reading MRP model results we need to remain aware that:

a) the methodology is relatively new and largely unproven

b) the range of results from different providers is wide (the Reform Party for instance is predicted anything from zero to 19 seats depending on which one you read)

c) the margin of error on any single model is also wide (always diligently called out by the pollsters, but naturally absent from the headlines)

d) despite the large sample sizes MRP models rely even more than normal polls on statistical modelling and therefore what they report is at least as much about the input assumptions as about the actual survey results. 

Caveat lector!

Coming back to YouGov - I don't for a second think that the share price movement has to do with accuracy of polling. As discussed, political polling is a small fraction of overall revenues. The profit warning is more likely driven by big brand clients' budgets being squeezed. Although in the US and UK we may not technically be in recession, and inflation is coming down, outside the tech sector growth is still patchy and costs continue to rise. In that environment marketing budgets are easy targets for profit-seeking CFOs.

As a former executive and current shareholder I expect YouGov to be at the forefront of adapting to the changes in media consumption and society, and the current share price could prove to be a buying opportunity.