The economy and financial markets are dynamic and evolutionary systems. Macroeconomic and market variables go through periods of stability and instability – often in a cyclical pattern.

Asset Management firms, in particular, need to adjust their portfolios allocations proactively instead of reacting to changing market conditions. To do that, they need a systematic way to identify the current phase of a market cycle, shortlist the key drivers and select strategies that will outperform in future phases.

This task is incredibly complex.

To handle this complexity, some innovative and quantitatively savvy managers are turning to machine intelligence. Machine intelligence is uniquely qualified to address this problem because it is able to derive patterns across hundreds of asset price changes simultaneously – without the requirement to iteratively present hypotheses or to make the thousands of subtle connections required to develop unique, uncorrelated insight.

The image below represents 25 years of market and macroeconomic data across a set of macroeconomic time-series, including original and transformed features such as the yield curve, inventory indices, volatility levels and labor market utilization. Each node (dot) represents a set of similar time-periods and an edge (line) represents a connection between nodes – which is to say that proximity implies similarity.

Holistically there is a main circular shape, representing normal market progression – the economic cycle. Normal market behavior tends to have an upward sloping yield curve, exchange rates driven by supply and demand, inverse correlation between volatility and equities, cyclical patterns of unemployment, etc. The circular shape shows represents this normal progression between market states and represents a natural way to segment the circle based on desired level of coarseness.

What’s really interesting about this picture are the disjointed islands.

The time-periods contained in the top right disjointed island is shown along with the most significant statistical drivers of this time-period. For example, characteristics of this period include higher volatility, lower lending, lower capital utilization. While emblematic of this time period, these variable represent a fraction of those needed to understand the behavior on a systematic basis (reach out to us if you would like to see the full list).

A savvy and informed market observer would likely be able to identify this time period without the dates but they represent the credit crisis of 2008. These periods are disconnected from the other market dates precisely because there are no historical analogues. It stands to reason this was an unprecedented credit crisis, one that brought down Lehman Brothers and resulted in extraordinary economic dislocation.

The key is to understand when you have left normalcy and what is likely to happen next. This is why our clients run periodic refreshes of similarity maps to capture the market state evolution when they operationalize machine intelligence within their organization.

Refreshing your similarity maps frequently as market states evolve enables market participants to prepare for or even anticipate sudden and unforeseen changes such as oil drops, equity index volatility and commodity collapses.

Another way to visualize this is over time.

What we have done here is to generate a series of similarity maps programmatically.

Frame 1: 1/1990 --- 1/2000

Frame 2: 1/1990 --- 2/2000

Frame 3: 1/1990 --- 3/2000

…

Frame 180: 1/1990 – 3/2015

https://www.youtube.com/embed/bdnWE93Inf0

From this video, there are three main takeaways.

• The cyclic shape gets more defined in later frames as there are more historical analogues for market states
• The density of edge connections (reflected in image “jumpiness”) shifts periodically due to the stochastic (random) nature of market data.
• Finally, disconnected islands have varying concentration levels which is consistent with the relative stability and the existence of historical analogues

Why does this matter?

It matters because as noted above, in order to effectively manage your portfolio risk, you need to be able to dynamically adjust your portfolio allocation proactively to changing market states. Firms that are using this approach have an asymmetrical information advantage.

Finally, this approach has applicability across a broad range of financial services use-cases. A great example is Risk Management functions at large banks. They are able to fuse market state information with internal datasets such as borrower characteristics, instrument valuations and analyst ratings to reliably predict ratings transitions, defaults and gauge regulatory exposure.

The economy and financial markets are dynamic and evolutionary systems. Macroeconomic and market variables go through periods of stability and instability – often in a cyclical pattern.

Asset Management firms, in particular, need to adjust their portfolios allocations proactively instead of reacting to changing market conditions. To do that, they need a systematic way to identify the current phase of a market cycle, shortlist the key drivers and select strategies that will outperform in future phases.

This task is incredibly complex.

To handle this complexity, some innovative and quantitatively savvy managers are turning to machine intelligence. Machine intelligence is uniquely qualified to address this problem because it is able to derive patterns across hundreds of asset price changes simultaneously – without the requirement to iteratively present hypotheses or to make the thousands of subtle connections required to develop unique, uncorrelated insight.

The image below represents 25 years of market and macroeconomic data across a set of macroeconomic time-series, including original and transformed features such as the yield curve, inventory indices, volatility levels and labor market utilization. Each node (dot) represents a set of similar time-periods and an edge (line) represents a connection between nodes – which is to say that proximity implies similarity.

Holistically there is a main circular shape, representing normal market progression – the economic cycle. Normal market behavior tends to have an upward sloping yield curve, exchange rates driven by supply and demand, inverse correlation between volatility and equities, cyclical patterns of unemployment, etc. The circular shape shows represents this normal progression between market states and represents a natural way to segment the circle based on desired level of coarseness.

What’s really interesting about this picture are the disjointed islands.

The time-periods contained in the top right disjointed island is shown along with the most significant statistical drivers of this time-period. For example, characteristics of this period include higher volatility, lower lending, lower capital utilization. While emblematic of this time period, these variable represent a fraction of those needed to understand the behavior on a systematic basis (reach out to us if you would like to see the full list).

A savvy and informed market observer would likely be able to identify this time period without the dates but they represent the credit crisis of 2008. These periods are disconnected from the other market dates precisely because there are no historical analogues. It stands to reason this was an unprecedented credit crisis, one that brought down Lehman Brothers and resulted in extraordinary economic dislocation.

The key is to understand when you have left normalcy and what is likely to happen next. This is why our clients run periodic refreshes of similarity maps to capture the market state evolution when they operationalize machine intelligence within their organization.

Refreshing your similarity maps frequently as market states evolve enables market participants to prepare for or even anticipate sudden and unforeseen changes such as oil drops, equity index volatility and commodity collapses.

Another way to visualize this is over time.

What we have done here is to generate a series of similarity maps programmatically.

Frame 1: 1/1990 --- 1/2000

Frame 2: 1/1990 --- 2/2000

Frame 3: 1/1990 --- 3/2000

…

Frame 180: 1/1990 – 3/2015

https://www.youtube.com/embed/bdnWE93Inf0

From this video, there are three main takeaways.

• The cyclic shape gets more defined in later frames as there are more historical analogues for market states
• The density of edge connections (reflected in image “jumpiness”) shifts periodically due to the stochastic (random) nature of market data.
• Finally, disconnected islands have varying concentration levels which is consistent with the relative stability and the existence of historical analogues

Why does this matter?

It matters because as noted above, in order to effectively manage your portfolio risk, you need to be able to dynamically adjust your portfolio allocation proactively to changing market states. Firms that are using this approach have an asymmetrical information advantage.

Finally, this approach has applicability across a broad range of financial services use-cases. A great example is Risk Management functions at large banks. They are able to fuse market state information with internal datasets such as borrower characteristics, instrument valuations and analyst ratings to reliably predict ratings transitions, defaults and gauge regulatory exposure.