Citi Readies Price-Shock Detection System 

Citibank’s corporate venture arm, Citi Ventures, is close to completing a proof-of-concept on a bot that provides event data from social media platforms that affects commodities prices. 

The Price Shock Detection System (PSDS) bot was built by D10X, or Discover 10 Times, Citi Ventures’ internal program looking at new solutions for the bank and its customers. It uses artificial intelligence to sort and classify signals on a variety of features to present traders with information before mainstream news sources write about those events. 

The PSDS bot aims to be a decision-support tool that helps traders deal with all the other sources of data that potentially impact markets before traditional sources such as Bloomberg, the Associated Press, or Reuters confirm it. These inputs can be pulled from social media, as well as microblogs and chat forums.

Agrim Singh, hacker-in-residence at Citi Ventures D10X in Asia-Pacific, says one example of this is when US President Donald Trump tweets. “The moment he tweeted about him not looking at any stimulus for the US, the markets went down. You don’t need to wait for someone to write an article and for something to happen for the market to move. The market moved purely because of him tweeting,” he says.

There are other events that may not be as significant as Trump tweeting. Perhaps a dam in Brazil collapses, and trading routes are blocked. It could happen that an English-speaking news outlet only reports on this later, as news of the collapse is first covered by the local media in Portugese. But the time between the event and when it gets reported globally could be enough for markets to move significantly. 

“The question was, ‘How do we capture alternate sources of data and extract trends and developing events from them, and present them in a way that is actionable so that you get to be ahead of the curve long before someone confirms the news?’ The added angle there was that this might be a rumor and there’s a reason [a major news outlet] hasn’t posted about it because they’re trying to confirm the facts. But markets may move regardless,” he says. 

This is where the PSDS bot comes in. The bot catches “supply-or-demand-side shocks” as it essentially listens for potentially market-moving chatter on the internet. For this, the team at D10X is ingesting data from the likes of Twitter, Reddit, and other microblogs. 

At the moment, the use case PSDS focuses on are for oil traders, but Singh tells WatersTechnology that the bot has also received interest from Citi’s equities trading team. 

The bot was built during Symphony’s APAC hackathon in September, where it won in the “Work From Home Workflow” category and will progress to the global Symphony Hackathon Olympics in 2021, competing against other contenders around the globe.

But, work on the bot is “far from over,” says Singh. “The thing that we were working on initially was, ‘Can we even do this with social media data—is this possible?’ The answer is, ‘Yes.’ We fiddled around with a combination of which social media platforms are important, and it turns out platforms like Twitter do tend to be more important just because of the nature of the platform. People don’t post market-moving information on Facebook and other places as much as they post on Twitter, which is established as a platform for breaking news and information,” he says. 

That was part one. Part two, which D10X focused on during the hackathon, was the user experience. “Again, [as many traders are working from home], they don’t have the 10 screens that they have at work. So, we are further limited by the user experience that we can have when it comes to receiving this information,” he says. 

During the hackathon, the team tested the bot with data from public sources to see if they could create a realistic user experience. If that were possible, the bot would be able to do even more using Citi’s data. “The hackathon helped us to validate a lot of our approach around the user experience, which we’re now working on,” he says.

Singh and the team are now working to incorporate Citi’s data and build the bot out. After that, D10X will put the bot in front of traders and get feedback. 

“Then it becomes an iterative process; nothing goes from zero to production. A large part of it would be, ‘Here’s what we built,’ and then the testing and discovery process continues, traders will give us feedback, we’ll incorporate it, make some changes, and put it out again. This sort of cycle goes on until all the nuts and bolts seem in place. Then, we push this out to the production level where everyone in the [oil trading] team can use it. Then we start scaling it beyond just oil traders, because this applies to other desks and other traders,” he says.

The downside of trawling through social media and microblog platforms is the amount of noise that needs filtering out. This is where working with the traders to refine the bot helps. 

“We showed during the hackathon that anyone using [the] Symphony bot can define, say, a set of keywords and events that they want to focus on. And as those events start developing or trending, they’ll start getting pushed. That way, it allows you to cut through the noise. Additionally, traders are also able to tell us if any notifications are irrelevant so that future iterations of the system will not show those kind of events.” 

Traders will also be able to define how many items they want to be pushed in a notification, or how often they want to be notified, depending on their preferences.

Singh says a large part of the D10X work involves exploring and finding out if these problems are even real problems, or is it that I think that’s the problem, and I want to solve it.