How We Use Algorithms To Discern Market Bias
Is Inflation Really A Risk?
This edition illustrates how we test market biases, through our self-learning framework (example covered here). To highlight and contrast cross asset-class behaviour, we illustrate with examples from Equity and Credit markets. The edition concludes with some thoughts on algorithmic models and their utility in thinking about real-world phenomena.
As economies emerge from COVID lockdowns, market attention has landed on Inflation Lift-off. A section of market participants, and mainstream media, hold the view that this is the sell-off/top trigger for markets.
Is Inflation really a Risk?
Source for the data used in this edition.
The Inflation Effect?
Unlike natural sciences, where independent and dependent variables are bound in a neat relationship, financial markets are a social science laboratory. Prices seldom respond to one event stimulus in isolation. Ceteris paribus (all other things being equal) seldom holds in the real world. Often, multiple event stimuli impinge on price behaviour.
Our self-learning algorithmic framework is built along these lines:
- Each market-implied event stimuli is triangulated in isolation; testing the individual effect on the influenced variable (Equity/Credit market performance).
- This process is repeated by progressively incorporating additional event stimuli.
- Inflation tends to exert its influence on markets over longer timeframes (this and below). The framework repeats the exercise over multiple lead/lag time horizons. This captures the timing effect (we are generally most interested in the long-term).
Implied Inflation & Credit
Implied inflation bears a negative relationship with subsequent price performance for Credit (proxy: CCC & Lower US High Yield index; a proxy for the riskiest slice of the Credit market). The relationship manifests over longer timeframes. The primary monetary regimes in this period are highlighted.
Implied Inflation & Equity
Similar relationship with Equity (proxy: Russell 2000 index).
Bringing it together
Finally, we test for the Inflation Effect on future asset prices, by embellishing with two additional market-implied variables. The additional variables improve the model.
We prefer market-implied stimuli, as it represents the market’s collective expectations, compared to event stimuli that may be susceptible to gaming (e.g., printed Inflation, GDP).
Model Variables
- Financial Stress Indicator: An index that measures the degree of financial stress in markets. Zero = normal financial market conditions. Values below zero = below-average financial market stress. Values above zero = above-average financial market stress.
- Implied Inflation expectations over 10 years. These are derived from nominal and inflation-indexed bond yields of equivalent maturity.
- Nominal 10-year Yield.
- Window: A time switch that triangulates the optimal timeframe for the model.
The Model outcome is illustrated on Equity, and Credit markets. For contrast, model behaviour with AAA Credit (lowest risk Credit) is also shown below. Period: 2003 — present.
Hypothesis: Influencer variables have no effect on asset prices.
The 3-variable model is statistically significant across both Credit and Equity markets. The model highlights some interesting takeaways.
- Both Credit and Equity forward price performance are positively related to Financial Stress Indicator. A high value today (“high stress” market regime) is associated with a positive return over the subsequent specified time window.
- Both Credit and Equity are negatively related to Implied Inflation. However, lowest-rated Credit (CCC) is more sensitive to Implied Inflation than Equity markets.
- Nominal Yields have a contrasting effect. They are positively related to Credit, but negatively related to Equity.
- Overall Model Fit is better for Credit compared to Equity.
Intuitively; Financial Stress, Inflation expectations and Bond Yields may be expected to have the strongest impact on the lowest-rated Credit universe. The model corroborates this intuition. This is one the reasons we focus intently on the lowest-rated credit universe. Model behaviour is the weakest for the AAA Credit universe: a risk-off asset class; with Equity falling somewhere in between.
Model behaviour on CCC-rated Credit
We look at how model predicted values compare with reality.
Model behaviour on Equity
The enhanced 3-variable model does a reasonable job of predicting outcomes over the long-run; compared to individual variable models. It tends to model Credit better than Equity.
Point Predictions v/s Regime Think
Many investors build mathematical models and expect a precise prediction of real-world dynamics. Natural science solutions applied to social science phenomena are a tricky affair. As markets are made up of thinking (and often, unthinking) participants, uncertainty is the rule. Instead of expecting precise (and accurate) forecasts, we prefer looking at prevailing reality through the lens of Regime behaviour.
One of the ways we use our self-learning framework is to use models as a guide, and derive value from mapping prevailing market conditions on broad Regimes.
We illustrate an instance of this below. We depict a Markov Switching model on Financial Stress Indicator, overlaid on Credit and Equity prices.
High Stress regimes are typically accompanied by sharp sell-off in both Credit and Equity. The framework dynamically switches Regimes based on evolving market realities. Acting as a flag on changing market conditions, and feeding back into positioning at any point in time.
The current Regime is signaling a long equity, short CCC credit positioning; over a 3-year+ timeframe.
