Improving Risk Metrics; the Obsidian way

Introduction:

Financial Markets are inherently unpredictable and sometimes irrational; off-hand tweets by leaders can precipitate global selloffs that last a day but are then forgotten soon afterwards. It is naive to think that divining the chaotic market structure can be achieved simply through an algorithm. This belief alone is responsible for many of the spectacular failures of financial advisory firms who employ, often lavishly, Machine Learning technology.

Obsidian’s technology avoids this pitfall by maintaining a sophisticated ecosystem that evolves, learns and adapts to market behaviour by exploiting hidden patterns that are impossible to detect with the techniques used for image recognition. Instead of using a specific set of rules, technical and fundamental analysis (which may be applicable), off the shelf solutions in Python or any other high-level programming language, we use an enhanced generalised learning base which is more like an industrial assembly line than an Artificial Intelligence black box.

Being able to draw on a long history of the real market performance of our ‘component parts’ in asset and risk management, we have been able to discriminate as to what constitutes a good strategy within our framework.

The continuous feedback of the performance of the components nourishes a vast metadata within our ecosystem that is used in the transmission of knowledge. The result is a methodical and fully automated application of unsupervised machine learning and evolutionary computation.

Through this unique infrastructure Obsidian’s technology can dramatically improve returns while suppressing volatility and improve risk metrics, such as the Sharpe Ratio, for any portfolio.

The graphs below illustrate the improvements in Returns, Volatility and ultimately Sharpe Ratio using Portfolios containing representatives of benchmark portfolios and various ETFs.

Enhanced Returns:

The graph shows the ratio of the annual rolling Profit Factor for the Obsidian-DJI, Obsidian-NDX and Obsidian-Russell over the appropriate Index benchmark:

Where:

The graph includes a line to indicate the neutral performance where Profit Factor = 1.0

Take away points:

• A Profit Factor of say 2.0 means that for every dollar lost during a given period, a portfolio earns twice as much.

• A high ratio of Profit Factors means that Obsidian’s Portfolio outperforms the Index portfolio by that ratio; a Profit Factor Ratio of say 2.0 means that, in a given period, Obsidian’s Portfolio generates net profits twice as quickly against its losses as does the Index Portfolio.

• Over the period of the study the Ratio rarely fell below 1.0 which means that for the most part Obsidian’s Portfolios outperformed their benchmarks and delivered better returns.

• The out-performance was stable for the NDX and DJI (until early 2018) but more volatile than for the Russell; the latter could be because the Russell generally has lower returns and volatility then its counterparts.

• All asset allocation, rotation and re-balancing related trading activities that result from the application of our technology to create Obsidian’s portfolios are fully automated, from inception to execution.

Suppressed Volatility:

The graph below shows the ratio of the annual rolling realised volatility for the Obsidian-DJI, Obsidian-NDX and Obsidian-Russell over the appropriate Index benchmark:

Take away points:

• The realised volatility measures the degree of variability in the returns of a portfolio; the analysis uses the realised volatility over contiguous one-year periods.

• The Ratio of the Volatilities indicates the degree to which Obsidian’s Portfolios were more of less volatile than the Index Portfolios; Ratios of less than 1.0 demonstrate volatility suppression.

• The degree of reduction is stable over the period of study for the Obsidian-Russell (~55%) and Obsidian-NDX (~25%) portfolios which suggests Obsidian consistently reduces realised volatility.

• The degree or reduction is less for the Obsidian-DJI Portfolio, with an average of ~20% but with a significant spike in early 2018. However, it’s worth noting that this was caused by a positive move for the Obsidian-DJI (so ‘good’ volatility).

• The suppression of volatility over different time frames and bull, bear and neutral market conditions, demonstrate the resilience of our methods.

Improved Sharpe Ratio:

The graph below shows the ratio of the rolling Sharpe Ratios for the Obsidian-DJI, Obsidian-NDX and Obsidian-Russell over the appropriate Index benchmark:

Take away points:

• The Sharpe Ratio is a useful way to normalise the performance of portfolios; it demonstrates the degree to which a Portfolio pays back (Average Return) for the risk taken (Volatility of Returns), so for a given Volatility a higher Sharpe is preferred.

• The Ratio of Sharpe Returns demonstrates the degree to which Obsidian’s Portfolios are better at using risk that the Index Portfolios.

• The graph illustrates the degree to which Obsidian improves the return on risk across indices; a Ratio of say 2.0 means that for given level of risk Obsidian delivers twice the returns of the Index Portfolio.

• The Ratio is stable for the Obsidian-NDX (average ~1.7) over the whole and for the Obsidian-DJI (average ~2.0) from early 2015 onwards.

• The Ratio is more volatile for the Obsidian-Russell but always remains above 2.5 which means that its Sharpe Ratio is always at least 2.5x that of the Russell over the period of analysis.

• The inherent flaw in the Sharpe Ratio, namely that it treats positive moves the same way as negative moves and therefore penalises positive surprises, can be overcome using metrics such as the Sortino Ratio, but the results above will be similar.

Conclusions:

With a sound framework for Financial Machine Learning, Obsidian’s Portfolio Management System can consistently improve on Returns, reduce realised Volatility and therefore improve risk metrics such as the Sharpe Ratio.

Federico M Dominguez and Faisal Khan (Oct 2019)