Since Twitter went public in November, 2013, they have reported key performance metrics each quarter. For social media – Twitter, Facebook, and LinkedIn – a key performance metric is monthly active users (MAU). MAU, different from the total user base, is how many unique people have logged in that month. This gives an idea of how many people are actually using the network, instead of how many people have simply registered on the network.
In Social Media companies, MAU is an indicator of revenue potential: a key metric for investors. The more users, the greater the monetization potential. This, coupled with how much money per user Twitter can realize, are the two main factors determining the total revenue. In a publicly traded company, MAU and ad-dollars per user are expected either to keep pace or to grow following historical trends.
Twitter has released two quarterly reports since going public. Each time, while earnings were up, the MAU growth was considered disappointing. In the most recent quarter, MAU for the previous quarter grew 25% to 255M users from the previous year, down from the 30% growth of the previous quarter. As a result, the stock dropped 10% in after-hours trading. Clearly, MAU is an important metric.
Twitter is expected to report its next quarterly earnings on July 29, 2014 and analysts will be looking closely at the growth of Twitter’s user base. Will the growth continue to slide, or has Twitter figured out how to recover and retain users? In anticipation of the quarterly announcement, I have analyzed data from Twitter over the past year to measure what their quarterly MAU will be.
Twitter defines Quarterly MAU as “Twitter users who have logged in and access Twitter through our website, mobile website, desktop or mobile applications, SMS or registered thirdparty applications or websites in the 30-day period ending on the date of measurement.” As a proxy measure for logins, I use posting activity. While not every Twitter user posts, the percentage of Twitter users who do post is likely fixed. Growth in posting activity quarter over quarter is proportional to growth in logins, and hence MAU.
To measure posting activity, I created a sample of 6,355 accounts that posted at least once between July 1, 2013 and June 30, 2014. I chose 6,355 as the sample size since this would provide me with the MAU count accurate to 1% with a 95% confidence, which means that I would likely measure the same number 19 out of 20 times. During the period of study, I created 30-day windows and counted how many accounts in my sample tweeted during that time. I averaged the number of accounts tweeting in each time window where the period ended within the quarter to get an average number of accounts tweeting in a given quarter.
Since my raw numbers are just a sample of the Twitter population, I needed to determine the factor to scale from my sample to the actual MAU reported by Twitter. To do this, I used a second order polynomial model to fit my measured data to Twitter’s reported MAU using a least squares method.
Figure 1 shows the resulting forecast for the Twitter Quarterly MAU in the Quarter ending June 30, 2014. My forecast puts the MAU at 264.7±7.5M. This is a 21.4±3.4% year-over-year growth. The most optimistic measurement merely matches the 25% growth from the previous quarter, indicating that Twitter continues its slide in MAU growth. And there is a chance that the year-over-year growth will be as low as 18%.
Based on my measurements, it is unlikely that Twitter will have reversed the slowing of growth in the next quarterly call.
Was Earnings Call Based on Soccer Blip?
On Twitter’s earnings call, they reported 271M MAU as of June 30, 2014. This is on the high end of what my measurements predicted. My range was that the MAU would be between 257.2M and 272.2M. Even though the year-over-year growth is at 24% (less that last quarter’s year-over-year growth), Twitter stock increased at the announcement.
Many analysts attributed the strong showing to the FIFA World Cup, which ran from June 12, 2014 to July 13, 2014. The World Cup-related bump would explain why the actual number was at the top end of my forecast. My measurement looked at posting as a proxy for activity. Much of the World Cup MAU bump would have been passive: people logging onto Twitter for results without actually posting. This is an example of why proper measurement of experimental error is important when working with these kinds of forecasts. Neglecting the inheirent variability in the data would have led to incorrect results.
Featured Image courtesy of Wikimedia Commons, the free media repository
- K. White, J. Li, N. Japkowicz, “Sampling Online Social Networks Using Coupling From the Past,” 2012 IEEE 12th International Conference on Data Mining Workshop on Data Mining in Networks, pp. 266-272↵