Top 5 Pitfalls for Building Models

     

View the top issues modeler’s face

building_models_cover

The five things to avoid when big data analytics is applied to price optimization management for financial institutions.

Learn more about price optimization.

Using the wrong model form

using the wrong model form

A fundamental decision a modeler must make is which model form to use. No matter how sophisticated the statistical techniques one uses, if the model form is incorrect, then the results will not be good. It is important to use a model form that is consistent with the physical events that are being modeled.

Learn more about price optimization.

Not examining the data

Not examining the data

Before modeling, it is important to first look at the data and make sure it looks as intended. Remove any outliers (i.e., data points that are either incorrect or so far outside the norm that they skew the average).

Learn more about price optimization.

Past performance does not guarantee future results

Past performance does not guarantee future results

For any forecasting model based on historical data, there is the implicit assumption that previous behavior will continue into the future. This may not always be the case, especially there are large events in the market. To ensure your model does not get too stale, frequent model re-calibrations can be done.

Learn more about price optimization.

Confusing the signal for the noise

Confusing the signal for the noise

In any real life system, there is a lot of variability (i.e., noise). To ensure you are identifying true correlations the (i.e., signal), it is important to conduct out-of-sample testing.

Learn more about price optimization.

Balancing model sophistication and business interpretability

Balancing model sophistication and business interpretability

Adding sophistication to a model may improve results, but in the end, a model must be understood and used by people. Often times the user of the model is not the model developer. Ensuring the model can be easily interpreted may require less sophistication and lower performance, but it reduces the chances of model misuse.

Learn more about price optimization.

About The Author

Oren Lavie is a Senior Business Analyst at Nomis Solutions where he focuses on the Lending vertical. Oren works with Home Equity, Auto Loan, and Mortgage clients in the U.S. and Canada. Prior to working at Nomis, Oren worked for Citigroup in the Risk Management department. Oren has a B.A. in Mathematics from the University of Pennsylvania and a Masters of Engineering in Industrial Engineering and Operations Research from UC Berkeley.