The Importance of Social Media in Risk Management

By Lee Campbell

Background

The emergence of social media over the last ten years has been dramatic. The number of daily Tweets on Twitter is now around 500 million. Facebook has an even larger amount of daily activity and other sites like LinkedIn also generate an enormous amount of content. These channels provide a massive amount of live, continually updating data. Clearly social media is now a vital marketing medium and a great tool for reaching out to large audiences quickly and easily. However, how useful is social media in risk management, for example, in providing insight into market activity, trends and sentiment? Can it be captured and stored? And can the information obtained be organised into something meaningful to help make better trading and risk mitigation decisions.

The benefit of social media for marketing is well understood. The value of information in general is quite simple to quantify in a number of metrics, namely; the quantity of data, how fresh the data is and how relevant is the data to my business. In many respects, this data is a new, emerging commodity.

Opinion in trading is an important measure of what will happen. For example, if there is an overriding opinion that the there is too much oil being produced, then the price of oil will likely fall. Other factors can impact the price such as; political instability or unrest in a major oil producing country, or news of a specific cut in oil production in a particular region or within OPEC.

The same sort of news or opinion about a particular sector can impact the value of a currency on foreign exchange markets. Negative news or sentiment about a particular company could ultimately impact its credit worthiness. In other areas of trading, weather information can have an impact on commodity prices, as the need for heat or power consumption is driven by climate in most regions. All these indicators are likely to be present, if perhaps buried in a sea of information, on social media ahead of any change in market price of a particular commodity, financial instrument or CDS spread.

Modelling the data

Let’s first consider how fresh the data is. Of all the ways in which good data can turn bad, loss of relevance due to age is one of the most challenging and most frequently overlooked. It may have been perfectly accurate and useful at a certain point in time, but aspects are almost certain to change over time. This can happen to any type of data at any time. Trading firms spend vast amounts of money and time to receive real-time market price information. This is vital to making the right trading decisions.

In social media, ‘topic models’ are methods used to organise, categorise, and understand textual information. As we have seen in recent years, the amount of textual data available is vast, so applying a good model is vital in enabling us to find relevant patterns, which in turn can provide us with insight into what topics are being discussed and how much relevance it has to our specific area of interest.

One of the main challenges in making sense of social media in risk management is the sheer amount of data, and the broad nature of the subject matters covered. However, while this is ultimately a technical challenge, a larger sample of data has the potential to provide a more accurate picture of a particular subject matter. Opinion polls carried out around the time of elections are generally tiny subsets of the voting population and in many recent cases have proved to be inaccurate. When questioned directly, people tend to answer less freely. An opinion poll is also a single snapshot in time, hence the need to take multiple polls over time and analyse trends.

Social media provides a less pressured outlet for one’s opinions and the number of opinions on offer would generally be a much larger and more representative set than a simple survey, and it would be continually changing as other factors change in time.

If one could capture social media data continuously, the main challenges would be; how to organise the data, filter out the relevant content, manage the different languages, different spellings and acronyms, strip out # tags and other misleading links. Big data challenges require complex algorithms and language libraries in order to make sense of data and organise it appropriately for relevant content. Several steps would be taken to organise and clean each data item before we apply it to a model for analysis. Reducing dimensionality, that is, reducing the number of variables that a model would have to deal with, will reduce computation times but at the expense of the model’s accuracy.

Other pre-processing techniques would include; language detection, correction of spelling mistakes, stemming (to strip unnecessary suffixes), and the removal of stopwords (commonly used prepositions), as they do not give any extra information to the model about the Tweet or post. We would also need to remove emojis (symbols), and http text. For the model itself, matching specific key words with collections of positive or negative sentiment word matches would provide clear, relevant and up to date sentiment insights into a particular country, company, product or commodity. However, we will see later that this alone is only a very simplistic lexicon model that doesn’t consider the context in which the word was used.

Credit Risk

Once we have a set of organised, relevant data that matches our target subject matter, we can look to apply techniques to provide insights for Risk Managers. Risk Management involves analysing lots of information and data and then taking steps to mitigate the risks apparent in that data. Risk Management is about forecasting based on known facts to minimise losses. The availability and timeliness of information and data is vital in making these decisions.

Credit Risk Management could benefit greatly from sentiment and insight from an organised social media data set. If this data is streamed in real-time, filtered for relevance and absorbed into an advanced topic model, it can be used to provide sentiment and timely information for Credit Risk Managers to analyse and monitor particular counterparties. That could then proactively prevent the impact of future credit risk events on the portfolio. Newsfeed type data could be highly useful and give more insight into a particular counterparty than would otherwise only be available from general information gained from a previous year’s financial statement, or a current public rating agency.

Historical social media data could also be useful as part of a more detailed analysis of a counterparty when considering the on-boarding of a new counterparty or performing credit reviews to set up trading limits for an existing counterparty.

Consuming the information is vital in conveying a specific message or theme. Sentiment can be considered in terms of specific words or messages. Figure 1 shows an example word cloud visualisation for a specific counterparty, which is currently under scrutiny in the market. Other options might include; a revolving news feed as information is streamed and analysed through the model, or a dashboard of sentiment changes over time. Business Intelligence visualization tools can help us present the information in meaningful ways for different audiences.

Energy Risk Software

Figure 1 – Example Word Cloud

Market Risk

Market Risk Management is largely based on the availability and timeliness of accurate price curve information. This information is gathered from current bid/offer prices across the trading community. Live data from the market drives changes in the valuations of open positions which increases profit or loss. Market Risk Managers and Traders need up-to-date information to assess the impact of open positions and formulate trading or hedging strategies to reduce risks inherent in those positions.

Social media in risk management could provide timely insights into a particular market or commodity. If such data could be streamed, organised and presented quickly to a Trader or Risk Manager, more timely trading or hedging could take place. In many scenarios, the speed at which a trade or hedge takes place is vitally important. As news or information feeds across the trading landscape, prices will change accordingly and potential profit making or loss avoiding decisions need to be taken in advance of such changes to have maximum impact.

Social media information could also contain valuable insights into market events and geographical or climate related events which will in turn, drive changes to the price of a financial instrument or commodity. Looking at the geographical location of a specific Tweet or post could provide insights into specific activity in a particular region or pricing hub.

How can we gather this information and present it in a timely manner to mitigate risk, and improve trading decisions? Again, a good ‘topic model’, with live social media newsfeeds can complement information already provided from sources like Bloomberg, Reuters and other news channels. Social media data has the potential to be able to provide timelier information on sentiment before more mainstream media has been able to analyse, structure, edit and finally present a particular event or trend to the marketplace.

Trend Analysis and Machine Learning

Machine Learning and Artificial Intelligence (AI) as a technology is becoming more widespread and mainstream as computing power increases and machine algorithms become more sophisticated. Our captured and organised information could potentially be used in machine learning algorithms to predict future events, discover trends and provide trading recommendations ahead of changes in price curves. However, it’s important to understand that different models will provide different levels of insight based on their complexity and sophistication. The accuracy of a sentiment model will depend on the quality of the subjectivity lexicons which are pre-built in the library.

In other words, the number of negative and positive lexicons that have been defined has a major impact on the accuracy of the model. Another consideration is that this type of model is not a context-aware model; Tweets that have a sentence containing: “amazingly bad” may be catalogued as neutral as “amazingly” may be considered positive and “bad” negative. More advanced, context-aware models will ultimately provide better results.

When looking at the social media information regarding a particular country or counterparty over time, a sophisticated machine learning algorithm could be used to detect subtle changes in mood or sentiment that could otherwise be missed. A trend of slowly increasing negative sentiment might imply something could happen in the future. Hiring information could also be considered, for example by using information from LinkedIn, the analysis of the length of time a new senior management employee remained at a particular company might indicate all is not well at that firm.

Conclusion

It is clear that the relevance of big data in all areas of our lives has become very significant. The technology available to consume this data and make sense of it is becoming more mainstream and better understood. The challenge for consumers of this information is in how to organise the data in a meaningful way and then how to visualise it. How to filter out the noise to quickly get to the relevant trend or sentiment that is interesting to the daily business. In trading and risk management, the importance and use of such data is only just being understood.

If technology companies can harness social media in risk management to provide risk managers with greater insight and enable more proactive risk mitigation techniques, this will bring enormous benefit and dramatic change in the future. The emergence of machine learning, coupled with the availability of good, relevant information from social media sources, and the application of business intelligence visualisation and reporting, has the potential to change the face of risk management in the, not so distant future. We see this as an important part of the future business intelligence enabled risk management landscape.

About CubeLogic

CubeLogic is a leading provider of Business Intelligence enabled energy, commodity & finance risk management software for market, credit, liquidity and regulatory compliance risks.

CubeLogic was founded by industry veterans in 2009, with a clear vision to provide out of the box risk and reporting solutions using the latest Business Intelligence platforms and concepts. Through rapid, global growth, we now serve major tier one organisations within the energy, commodities and financial services sectors. We continue to provide value to our clients with offices in the UK, USA and Bangalore.