Factor Risk Model Returns Timing Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current factor risk models fail to accurately estimate the risk of investment portfolios with assets traded on non-uniform or overlapping markets, leading to distorted correlations and volatility predictions due to asynchronous trading hours, particularly affecting American Depository Receipts (ADRs) and global asset models.
Innovation Solution
A vector auto-regressive (VAR) model is used to adjust factor returns and correct for the effects of asynchronous trading, allowing for more realistic correlation modeling across markets and improving the estimation of specific risk for ADRs and other assets, by synchronizing returns based on the behavior of markets with overlapping trading hours.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If factor risk models use standard daily returns from markets with different trading hours, then the model can be constructed simply, but the correlations across markets become distorted and risk estimation becomes inaccurate
Solution Approach 1:
The patent applies preliminary action by adjusting factor returns before they are used in the risk model. Specifically, it uses Vector Autoregressive (VAR) models to predict what market returns would have been if all markets traded simultaneously, then applies these predictions to adjust the actual returns. This pre-adjustment eliminates the distortion caused by asynchronous trading hours before the returns are fed into the factor risk model, thereby improving measurement precision without requiring complex real-time synchronization mechanisms.
Solution Approach 2:
The patent introduces an intermediary mechanism - the VAR model - that mediates between the actual asynchronous market returns and the factor risk model. The VAR model acts as a translator that converts returns from markets with different trading hours into a common reference framework, allowing the factor risk model to process them uniformly. This intermediary layer resolves the contradiction by providing a systematic method to handle market timing differences without fundamentally changing the factor risk model structure.
2Measurement precision
If ADR returns are calculated using closing prices from different trading sessions, then data collection is straightforward, but spurious autocorrelation is introduced and specific risk tracking is impaired
Solution Approach 1:
The patent applies preliminary action by adjusting ADR returns before they are used in specific risk calculations. It uses the VAR model to predict the underlying market returns that would have occurred during the ADR trading session, then uses these predictions to adjust the ADR returns. This pre-adjustment removes the spurious autocorrelation introduced by asynchronous trading before the returns are used to calculate specific risk, thereby improving measurement precision while maintaining data collection simplicity.
3Measurement precision
If markets with non-overlapping trading hours are modeled without adjustment, then the model structure remains simple, but volatility predictions are distorted
Solution Approach 1:
The patent applies parameter changes by transforming the return parameters to account for different trading hours. Instead of changing the fundamental model structure, it adjusts the return parameters using VAR predictions that incorporate the specific trading hour characteristics of each market. This allows the model to accurately capture volatility patterns across markets with non-overlapping hours while maintaining a relatively simple model framework.
Data Source
AI summary
Until recently, risk models have been built using low frequency data, such as weekly or monthly data. This approach has resulted in a necessary compromise between model stability for which one needs a long history of data, and model responsiveness, for which, the shorter the history, the better. Stability plus responsiveness can be achieved if one uses daily data, which allows for a large number of observations to be used in model estimation without using long out-of-date data. Daily data have other problems, however, as the differing closing times of markets worldwide may induce spurious relationships across model factors. In particular, correlations between markets may appear lower than they truly are due to a market lag effect. To address such issues, a stable, daily data-based factor risk model is described which takes account of the differing market closing times and corrects the model factor correlations and specific returns accordingly.