Analyzing KPIs

Why KPI analysis?

When using alternative data in a fundamental investment process, combining data sets to track and model company KPIs is crucial. Modeling serves to translate trends observed in alternative data into the actual financial / operational performance of a company. By combining data in this process, users fuse information from multiple sources to reduce prediction error. This allows real-time tracking of company KPIs that ultimately drive intrinsic value, and also helps to identify near-term trading opportunities where models are predicting a significant beat / miss versus consensus.

Pain points

This is a manual and painful process with traditional tools. Most vendor data is not mapped to relevant KPIs, or in a limited way to generic revenue KPIs, forcing the buy-side to do this themselves. Rigorous modeling and backtesting is hard to do in tools like Excel and requires significant expertise, putting it out of reach of non-technical analysts and PMs. Custom ad-hoc analysis is not repeatable and hard to share across teams, hindering the ability to use alternative data at scale across an organization. Existing solutions on the market provide little transparency into how vendor data has been mapped, and do not allow analysts to customize analysis according to their investment theses.

👍
Exabel's solution
Exabel solves these problem in 3 key areas:

Curated and customizable: Exabel has curated thousands of KPI mappings for integrated vendors, covering not just generic top-line revenue but also key company-specific KPIs. Users can build upon this by customizing mappings and models to suit their investment theses and create their own proprietary analysis.

Combining data: Exabel helps users combine all the data they have available to train the best possible model. “Auto models” automatically optimize the weights of each vendor signal for you, but there is also flexibility to build custom models with a particular choice of vendors.

Rigorous modeling: Exabel provides a production-grade model backtesting and training framework that is extremely performant and also easy for non-technical users to use.

KPI analysis in Exabel

Exabel breaks down KPI analysis into 3 steps: Mapping, Modelling and Monitoring.

Mapping

Mapping is the process of linking vendor data to a relevant company KPI, where there is a fundamental relationship between the data and the KPI, e.g. between card spending and company revenue. At its core, each KPI mapping is a relationship between 2 time series - the KPI and a “proxy”. This is configured by choosing a company, a KPI from FactSet or Visible Alpha, and a proxy signal. Because the proxy is defined with a signal expression, this allows for a large degree of creativity in how to best transform the data in order to more accurately track the KPI.

It is possible to create both bulk and company-specific mappings in Exabel. Bulk mappings take hundreds or thousands of companies and create standardized mappings to a common KPI (e.g. sales) and proxy signal, which helps to bootstrap a new data set. This is then complemented by company-specific mappings that target the most important KPIs for each company with fine-tuned proxy signals to optimize for predictive power.

Exabel provides curated mappings (both bulk and company-specific) for integrated data vendors, to ease this laborious step and help customers get started more quickly. Statistical analysis such as correlations and backtests are run automatically on every KPI mapping, which can then be compared side-by-side to identify the best proxies for each KPI. These results are available to view even without a vendor subscription, allowing for easy discovery of data sets that might be additive.

At the same time, users can customize mappings further in many dimensions:

Choose a different KPI that the market might not be paying attention to yet
Map to a different slice or subset of data, e.g. a specific set of web domains/pages
Transform the data in a different way using signal expressions

As users are testing new mappings, Exabel runs the same statistical analysis and backtesting within seconds, giving quick feedback on whether there is a statistically significant relationship with the KPI. Once an optimal mapping has been found, this can then be saved and used in the subsequent modeling step.

Modelling

Modeling is the step of taking one or more KPI mappings and combining them to train an optimized model that makes accurate predictions of the KPI. Such models often combine data from multiple vendors in order to reduce prediction error.

Exabel's KPI Analyzer offers a battle-tested, production-grade modelling framework:

Easily combine multiple sources/vendors, including curated vendor KPI mappings and your own data.
Choice of popular open-source machine learning models, as well as proprietary Exabel models fine-tuned for alternative data (white papers available on request).
Lightning fast: test new models in seconds, allowing you to quickly optimize for the best combination of data sources and model parameters.
Walk-forward, point-in-time backtesting: produces out-of-sample results and more accurate model error estimates that take into account where we are in a quarter (& thus how much data is available).
Daily refresh with revision tracking: model inference is run daily to take into account the latest data, and model revisions are automatically tracked to provide a log of how model predictions have changed over time.

Monitoring

Finally, models can be monitored regularly to track how companies are performing in real-time. Model predictions are refreshed daily with new data. Exabel’s KPI monitoring view provides a high-level view of model predictions across all KPIs for a company. Models can also be put on dashboards or set up with alerts.

All mappings and models can also be shared with other colleagues on the same team, or even broadly across the organization. This allows for collaboration and sharing of ideas and results, allowing a team to scale KPI analysis faster.

Getting started

KPI analysis in Exabel is performed through the KPI Analyzer and Mapper. Continue to learn more about how to perform your own KPI analysis.

KPI Analyzer: map, model & monitor KPIs on any company of interest
KPI Mapper: browse & manage KPI mappings, and review results across collections of KPI mappings