Regulatory Compliance and Data Management

Data traceability

Spotfire puts in the hands of business users the ability to combine any number of disparate data sources and wrangle it to find desired outputs. Spotfire^® keeps track of all calculation steps, such they are always traceable and, more importantly, repeatable. Being a very visual environment, results can include not just required data but metrics of data quality.

Figure 1. Example of the source view diagram of a Spotfire table. Every transformation step is dynamically kept track of and will be reapplied upon data refresh.

Timeliness of reporting

Once a report is built in Spotfire, opening it is all that is required to update it with the most recent information. Spotfire reconnects to all used data sources and applies precisely the same calculation steps, so users can see yesterday's report with now data with the click of a button.

Inbuilt sensitivity analysis and predictive analytics

Be it what-if analysis, Monte Carlo simulations to be run on a grid computing environment such as TIBCO GridServer, or advanced statistical models, Spotfire not only lets you see today's valuations but also what their value would be if their guiding parameters, e.g. risk-free interest rate, were different. Spotfire allows calling calculations in Spotfire Expression Language, TERR (TIBCO's R), open source R, SAS, Matlab, KNIME, Python, C++ via GridServer, H2O, or Spark. These calculations can be as simple or as complex as you require them to be. For example, it can be your credit manager inputting new market information about the specific credit rating of a company to see the impact that has on portfolio valuation. Or it can be your CEO sliding a bar to input his/her beliefs regarding the macro-environment: this value be passed into the valuation of all assets, recalculating their value differently per each asset category, and bringing the final result back to Spotfire for your CEO's appreciation. Check out this simple example of using Spotfire to measure operational risk, our Template on the Community site.

Figure 2. What-if analysis: how does a change in the scoring of a holding affect the portfolio risk exposure?

Figure 3. Monte Carlo simulation using Spotfire and TERR to simulate a scenario-based view of Economic Capital needs resulting from changes in the macroeconomic environment. The user specifies parameters that determine the joint distribution of these two risk factors and their contribution to Total Losses. Spotfire displays consequent estimated loss distribution of a dependent asset and respective economic capital needs (blue area of the bar chart).

Shared Best Practices and Increased Transparency

Spotfire reports and dashboards can be shared over intra or extranet. The same report updates with only the data the signed-in user is allowed to see. Your customers can share a report of the value of their holdings NOW just as your business, even internationally, can share dashboards that encompass best practices in any field, from assessing customer creditworthiness to investigating transactions or viewing economic capital requirements. And if the CEO logs in, he can get a view of the fully aggregated international position from the same dashboards your people used in their micro-decisions, drill-down to any levels of interesting behavior, comment, and drill back up. An international American bank found an increase of 80-90% in productivity best practice dashboards that can be aggregated natively from this aspect alone.

Figure 4. Sharing dashboards increase transparency and productivity. Notice how one dashboard can give a row-level or aggregated view.

Decision traceability

Customer protection is an important requirement of many FSI regulations. It requires being able to explain, for any given customer, the reason that drove the decision about them, e.g. denial of credit or credit card transaction or money transfer. In Figure 6., we can see that the customer or his transaction is considered dangerously similar to past known target cases, according to the supervised machine learning model with version number 82. The threshold for separating cases was then 90% and the customer has a 100% target probability. We can see which were the features of most relevance to the model and the customer's/transaction's position in those features, in this case, all in the red zone. 

Figure 5. Explaining a decision regarding a particular customer.