by Ben Hazzard
This is an example of how people are exploring data and AI for investing research. Ben analyses and explains how he’s using interesting datasets and AI tools to look more deeply into a restaurant/night bar chain: Revolution Bars (LON:RBG) .
Over the past week I have been taking my brother in law and his wife on a whistle-stop tour of the UK as they are visiting for the first time from Canada. I am in my early thirties and they are a little younger so we still very much indulge the city nightlife where we can. Normally I would take little interest in Revolution Bars (LON:RBG) but I have been contemplating a small position as a bit of a punt so kept my eyes open for them. In both Manchester and Liverpool it looked busy and their spin off bar Revolution de Cuba was heaving. It made me think how valuable it would be to know how all the sites around the country were trading in real time, especially compared to previous weeks, months and years.
This encouraged me to look a little deeper and I spent a bit of time over this weekend doing some analysis of the data that google collects on how busy an establishment is at any given point in time. Google’s Live Popular Times feature compares the current number of people at a location to the average number of people at that location during the same time of day on the same day of the week over the past few months.
For example, if Google says that a restaurant is busier than normal at 7pm on a Tuesday, it means that there are more people at the restaurant at 7pm on this Tuesday than there were at 7pm on Tuesdays in the past few months.
This data is collected from people’s mobile phone location data and some choose not to share this, so it does not give an entirely accurate data set nor exact numbers of visitors. It should however be quite precise when used a comparable value over time to assess how busy a bar is.
I collected a single data point from each of the 48 Revolution Bars locations and the 18 Revolution de Cuba Bars – Saturday 15th Match between 8 and 9 pm. What should be one of the busiest trading times of the week.
25/48 Revolution Bars sites were listed as “as busy as it gets”, 6/48 were “busier than usual”, 4/48 “a little busy”, 4/48 “not too busy”, 3/48 “less busy than usual”. 5 I could get no data on, and one was closed.
11/18 Revolution de Cuba sites were listed as “as busy as it gets”, 3/18 were “busier than usual”, 2/18 “a little busy”, 1/18 “less busy than usual” and 1/18 I could get no data on.
This to me seemed quite a positive reflection on their current trading, suggesting an improving picture in recent months. A contrast to the more negative prospects for hospitality predicted by our mainstream media. I decided to use Google’s Bard to see if I could do some further analysis. Bard was released last month and is similar to Chat GTP however it is a language model that can access and process information from the real world. This means I was able to ask Bard to carry out an analysis of all of the 2023 H1 busyness data, something that would take an age to analyse in the way I had just done.
It’s response was:
“According to the data from Google’s Live Popular Times, Revolution Bars has been more busy than usual across their UK bars in H1 2023. The average busyness of Revolution Bars across their UK bars in H1 2023 is 80%, which is higher than the average busyness of 70% for the same period in 2022”
I also got Bard to look at Instagram data, specifically on the number of times Revolution Bars group was tagged by people in their photos and posts:
“According to my knowledge, there have been over 102,000 tags at Revolution Bars Group locations in the first 3 months of 2023. Overall, the number of tags at Revolution Bars Group locations has increased significantly in the first 3 months of 2023. This is a significant increase from the same period in 2022, when there were only about 50,000 tags”
I also asked Bard how the number of tags and busyness data compared to other publicly listed bars groups:
“The increase in the number of tags and the busyness data at Revolution Bars Group locations compares favorably to other publicly listed bars groups. For example, Wetherspoons, which is the largest pub chain in the UK, has seen an increase in the number of tags at its locations in recent months. However, the increase at Revolution Bars Group locations has been more pronounced. This suggests that Revolution Bars Group is benefiting from a stronger recovery in demand than its competitors.”
This data may not be the most reliable and there are without doubt flaws in the approach, I recognise the busyness data and Instagram tag data may end up having very little correlation to the results when they are out. Despite this, I feel better informed than I did before this weekend about Revolution Bars current trading and to quote Paul “I am going to dip my toe in” having previously been on the fence. I hope a few of you found the process thought provoking.
Please note: I am concluding with my short experiments with Bard and ChatGPT that anytime they use numbers the results must be checked in detail. I surmise they have clearly not been trained in quants (quantitative) in any shape or form. That noted, the two platforms have been trained on vast volumes of data (but very different data sets, so each gives different responses to the same question), which is text, which is language. Thus they seem to be able to offer some quite useful pointers along the qual (qualitative) front. In short, I think the current iterations of Bard and ChatGPT can offer some support for investors who might be digging into the context of equities, markets and the financial domain. But the two AI platforms ought to be viewed as a rather flawed but fast working researching or data gathering assistant. Who must be supervised and checked closely.
