Reuters Conversation Analysis
When trading on Reuters Dealing 3000 there is always a risk that a valid trade will pass through unconfirmed. This means that the trade could be missed unless it is picked up during the checking process.
With many hundreds of deals passing through each day manual checking of conversations is both expensive and error prone. RCP Conversational Analysis automates this critical process with very high levels of accuracy irrespective of the volume of transactions being traded. This reduces staff costs significantly and greatly reduces the operational risk and reputational risks that go with a missed FX trade.
What is RCP Conversation Analysis?
RCP Conversation Analysis acts as a virtual Back Office staff member. Before any human review Conversation Analysis pre-sorts the Reuters unconfirmed conversations into three categories:
- “NO DEAL” conversations that do not contain an agreed trade.
- “QUERY” conversations that might contain an agreed trade.
- “DEAL” conversations that are highly likely to contain an agreed trade.
No matter what your Back Office process is, our experience has shown that it can be dramatically improved using RCP Conversation Analysis.
Typical Analysis Results
RCP has tested the latest version of Conversation Analysis (Version 2.52) on hundreds of thousands of Reuters conversations, provided by real banks. The typical results below are obtained from this data. These results are based on the assumption that traders fail to confirm 1% of agreed trades.
Out of 100,000 conversations typically:
• 39,660 conversations categorised as “CONFIRMED”
The deals agreed in these conversations have already been captured by the trader.
• 37,962 conversations categorised as “NO DEAL”
RCP Conversation Analysis has identified conversations as not containing a deal, using this sample data an accuracy of 99.998% was achieved. This means that it is unlikely any of these conversations would contain a deal.
• 929 conversation categorised as “DEAL”
Out of these 929 conversations, 385 will actually contain a deal.
• 21,449 conversations categorised as “QUERY”
Out of these 21,449 conversation, 12 will actually contain a deal
How much will RCP Conversation Analysis save?
Organisations that review all conversations: Only reviewing conversations in the “QUERY” or “DEAL” categories could achieve the same missed deal detection rate but would reduce Back Office workload by an amazing 77%. Alternatively the “NO DEAL” category is subdivided into deal probability, there is a diminishing effort to deal detection rate as the Back Office checks conversations that are assigned lower probability. The organisation can choose the level of probability they are comfortable to reject.
Organisations that sometimes need to review a backlog of conversations: RCP Conversation Analysis can order the conversation by the probability of it containing a deal, (e.g. 99% for textbook agreed trades, 1% for broadcast messages). When a backlog occurs missed trades can still be identified and actioned quickly reducing cost and risk.
Organisations that do not review any conversations: Due to the large costs associated with checking all conversations, some organisations have chosen to take the financial risk to only reviewing a sample of conversations. Using Conversation Analysis and reviewing only conversations categorised as “DEAL” these organisations would identify around 97% of missed trades, achieving significant reduction in risk for minimal cost.
About RCP Conversation Analysis
RCP’s experience developing Reuters Dealing 3000, 2000 and RDL helped us to understand the best way to analyse Reuters Conversations. The result was RCP Conversation Analysis:
a) Over 10 man years of development effort has been invested in the development of the RCP Conversation Analysis algorithm.
b) RCP Conversation Analysis is written using a standard programming language meaning there were no limits to the complexity of underlying algorithm. This is important as the flexibility of the logical language used to write any conversational analysis application is the main limitation on achievable accuracy.
c) RCP Conversation Analysis is continually upgraded to handle new trading styles and instruments. Before each upgrade is released it is fully regression tested using automated test scripts involving hundreds of thousands of deals to ensure that the accuracy of all analysis is maintained.
d) When any user requests an update to RCP Conversation Analysis, for example to capture a new trading style, RCP makes the updated module version available to all users. Thus the accuracy is improved for everyone.
