Financial Market Modeling

Where qualitative meets quantitative through AI

The following page will discuss how to optimize portfolio returns by analyzing data and modeling financial systems through deep learning neural networks.  Core concepts include GAAP Policy, MapReduce,  Exploratory Factor Analysis, & Principal Component Analysis. The aggregate investment data was generated using the Bloomberg terminal API.

Understanding The Economy

Where Things Fit Together

 The most critical component when creating a model is picking the best features or measurable properties of phenomena you are observing. To accurately model the financial market I considered the three underlying area's of study which governs it, accounting, economics and computational finance. Each area of study was represented in my model through meticulously designed features.

 

Accounting is a critical component which defines a company's fundamental structure. For instance, different accounting methods can drastically change a company's financial statements. Second, knowing a company's financial structure is important in assessing a companies default risk from a deleveraging.

Economics provides insight into forecasting company value with respect to industry performance. Understanding consumer spending and income provide insight into the underlying factors which drives company performance. Income and spending are approximated through three main mechanisms, interest rates, inflation, and the risk-free rate of return in the market. Note, a later project of mine introduces an American Census Data predictive model which provides data on hundreds of thousands of geospatial locations.

Computational Finance generally works best when dealing with somewhat stable markets. Several features or indicators used are Q learning, Holts Exponential Smoothing, volatility in the S&P Efficient Frontier, the predicted optimal ratio of stocks and bonds of a company derived from delta hedging when using Ito's lemma.

Baseline Vs. Portfolio Results

Sample Allocation

Portfolio Performance

Accuracy Rate = (True Positives + True Negatives) / Total Samples

Error Rate = (False Positives +  False Negatives) / Total Samples

Uncertain = Signal was to speculative to make an informed call

Portfolio Log Performance

© 2018 by Alex Geiger

  • Facebook Social Icon
  • LinkedIn Social Icon
  • Google+ Social Icon