Introducing the Implied Correlation Signal

Hi everyone, this is my first post on WealthUmbrella! I’m Zackary Stephenson, and I’ve been coding behind the scenes for WU since the beginning of 2024. Today, I’m here to introduce one of the indicators I’ve been working on recently. But before that, I thought I’d share a bit of my background.

During my engineering degree in Mechanics at Sherbrooke University, I quickly realized from the few courses on programming and robotics that this was the field I wanted to pursue for my career. From that moment on, I took every elective course related to robotics, AI and Vision and applied for internships in the field. That’s how I became an intern at Kinova Robotics, a leader in the field of medical robotics.

Working in the R&D department of a tech company convinced me that I wanted to push my studies further and pursue a master’s degree. In some fields of engineering, an undergraduate degree prepares you well for what comes next, but since robotics is at the crossroads of so many engineering disciplines, pursuing a graduate degree to specialize in one of these areas is almost the default path in this field.

After expressing my intentions to Kinova, I was pleasantly surprised when they offered to fund my master’s degree, provided I worked on a research project for them. For academic supervision, they suggested I contact Professor Duchaine at ETS due to his strong background in collaborative robotics, which I promptly did. Vincent agreed to be my supervisor, guiding me on a signal processing project that involved using six very noisy actuator signals to monitor potential failures of a multi-axis force/torque sensor at a robot’s wrist.

This might sound like a highly niche robotics project, but when you start working with the data, it actually has a lot in common with developing strategies for the stock market. After all, what we do at WU is use noisy data to detect changes in trends (sensor/stock market failure). In fact, I even used Bollinger Bands in the algorithm we developed during my master’s degree. But I’ll spare you the boring math details of that project—the essential thing is that this project, combined with the weekly meetings I had with Vincent, got me hooked on the quant world.

After finishing my thesis, I had a good discussion with Vincent about some work possibilities I had ahead. One thing led to another, and in January of this year, instead of working for a robotics company, I started coding for WealthUmbrella. I was also involved in another startup on the side, but I realized that what I loved most during my weeks was doing quant analysis for WU. This prompted me to leave that startup a month ago and join WU full-time, now as a co-owner. This probably means you’ll be seeing more of me in the coming months.

New Option-Related Indicator

One of the things I did recently was working on spinning off an indicator from raw data that WU was already using in some of its algorithms, including the Hedge. Vincent explained this in his post about WU’s novel S&P500 Risk Index, but I’ll explain it again here for future WU members who may read this post when they want to learn about our Implied Correlation Signal work.

Before even starting to code any WU indicators, Vincent and Jennifer spent a lot of time analyzing hundreds of datasets to find those that were forward-correlated with market movements. The word “forward” is important because correlation alone isn’t enough if we hope to get early signals. Let’s take an example that’s currently in the spotlight: the unemployment rate can be forward-correlated with a recession, as shown by the Sahm Rule indicator. On the other hand, creating a recession indicator based on GDP would also yield a correlated signal of recession if we consider the criterion of two consecutive quarters with negative growth. However, the issue is that this signal, although correlated, would come at least six months after the beginning of the recession, unlike with the Sahm rule.

Finding which data are relevant to S&P 500 movements from the vast data pool available was a monumental task, and sometimes we feel that just giving the names of our underlying data would already be like disclosing the ingredients of our Big Mac sauce. In some instances, saying what the indicator is tracking isn’t sufficient to point to a single dataset, but in the indicator we are presenting today, it is pretty straightforward. It is based on an options dataset we know to be powerful in anticipating market movement, so we decided to use it in our new WU S&P 500 Risk Index. This implied that we had to disclose the model to help people understand all the components behind this Risk Index.

Implied Correlation

The new indicator we are releasing today is based on Cboe Implied Correlation, one of the Cboe options data sets we believe is most useful after the VIX and the SKEW. Specifically, we use the Cboe COR1M and COR3M data feeds. But what is Implied Correlation? Here’s how Cboe explains it:

“Implied Correlation, a gauge of herd behavior, is the market’s expectation of future diversification benefits. It measures the average expected correlation between the top 50 stocks in the SPX index. Cboe calculates COR3M by using ATM delta relative constant maturity SPX index and component option implied volatilities. […] The Cboe Implied Correlation Index measures correlation market expectations by quantifying the spread between the SPX index implied volatility and the average single-stock basket component implied volatility.”

Cboe sometimes makes it more complicated than it needs to be. In simple terms, they look at the difference between the volatility of the index (SPX) and that of the individual stocks within the index. But why might the sum of individual stock volatilities differ from the volatility of the index since an index is just the sum of its components? This holds true for price action, but not necessarily for volatility. Let’s take an example of a hypothetical two-stock index (S&P2) consisting of Nvidia and Berkshire Hathaway. Both could have a certain level of volatility of ±1% per day, but if one goes up while the other goes down, we could see our S&P 2 index moving sideways with very low volatility (let’s say ±0.1%). In that case, the spread between our Index volatility and that of its components would be large. But as we enter a period of risk, both could start moving more in sync, like going down together, which is usually what happens in a real pullback. With the same ±1% individual volatility, this would now result in an index volatility that closely matches this 1%, leading to a very small spread between the index volatility and that of the individual stocks (high correlation). So, examining the correlation between the volatility of the index’s individual components and the resulting index’s volatility can tell us how in harmony the stocks are moving.

But what does this volatility mismatch tell us? This dataset can be used for various purposes. For example, portfolio managers use it to estimate their diversification risk. From a market perspective, these data are very insightful. A very low or high COR1M usually indicates poor market breadth. If it’s very low, you have poor market breadth but an index in an uptrend, and if it’s high, you have a market in a downtrend. Would it surprise you if I told you that this index made an all-time low this year, as the index was driven by only a few companies (low correlation between the index volatility and that of its internal stocks)?

Since this data is somewhat related to market breadth, one might wonder why we need it in addition to our market breadth-related indicator. The answer lies in the fact that this indicator is not calculated using the actual realized volatility (the real experienced volatility) of the stocks, as in the example above, but rather is derived from implied volatility in a methodology similar to that used in VIX calculations. So, instead of reflecting the current price action, it is based on what option traders are anticipating for the future. This makes it more noisy than market breadth, but it also allows it to provide some lead time over a market breadth-related indicator. Additionally, as we believe redundancy is key, we think that examining the market from every angle increases our chances of flagging a pullback early.

The Indicator

The indicator is based on the first derivative of the CBOE COR1M and COR3M, which respectively track At The Money (ATM) option contracts with expiration dates within 1 month and 3 months. These two datasets have their own trajectories, offering two different points of view.

The way I combined them is not trivial to explain, and I doubt explaining it will really help, but our WU SP500 Implied Correlation Signal is the result of that integration. Whenever both the COR1M and COR3M reach their thresholds, our indicator will flip red. This usually happens around the red horizontal line we placed in our indicator, but it is not always perfectly the case.

The result of this indicator is a relatively strong signal that will usually ignore small dips. It tends to trigger early but will very rarely activate before a correction truly begins.

We didn’t upload this indicator separately in TradingView to keep low the number of signals we offer. To access it, you’ll need to open our new WU S&P500 Risk Index and go into the parameters to activate it. Once you're in the parameters, you’ll see this text box:

Simply select "Implied Correlation Signal".

Results

The statistics for this signal are relatively strong considering how simple it is, which tends to prove that the movement of recorrelation between stocks and the index is a sign of fear that usually culminates in bearish price action. In fact, with a success rate of around 70% over 47 trades from 2017 to now, and a return that outperforms buying and holding the index by 169%, this signal is in the same league as the options model, although it doesn’t trigger at the same time. Here are the complete stats for that period:

Conclusion

Implied Correlation is a signal that is highly correlated with market movements and examines the market from a different angle than our other indicators. Since markets never start their moves in the same way, we believe it is important to look at the market through different lenses. This new indicator, which we kept confidential for a while, is a great tool that allows us to flag environments where the market will usually experience significant declines. In that sense, we think it is a valuable addition, particularly as a component of our new WU SP500 Risk Index.