WU SP500 Hedging Signal

Protecting the gains made during a bull run is usually one of the things that hedge funds do that most small investors don't. But once you realize that it takes a run-up of 100% to recover from a 50% loss, you understand the importance of Warren Buffett's rule no. 1: never lose money.

There are plenty of ways to hedge a portfolio, including diversification into contrarian assets, volatility protection, and others. Hedge funds are masters at this and will constantly rotate to position a part of their portfolio into a defensive posture. One of the simplest ways of protecting our gains is going in and out of the market or shorting a correlated stock (or basket of stocks) that will go in the exact opposite direction from their portfolio in order to neutralize it. Inverse ETFs have also been created for that purpose (more on this topic in the last two sections of this book). In fact, there are infinite solutions to make a portfolio neutral in order to sail calmly through a storm, but most of these solutions rely on a well-defined and backtested signal that will tell when it's better to protect our portfolio.

What’s under the hood to the signal ?

Our hedging signal is the result of almost two years of work dedicated to finding a way to identify which market conditions lead to real deep corrections in order to give ourselves a signal to protect our capital. As we said before, we are more long-term investors rather than traders, so our challenge was to create an algorithm that keeps us in the market by ignoring most of the little drops along the road, and only flagging the real ones. This is not a trading strategy but rather our own weather forecast of the general market.

Like most people, we initially started working on that problem by building a system using the crossing point of different exponential moving averages. This strategy, very common among investors, was for us inspired by Puru Saxena, who has a very defensive style since he mostly invests in growth stocks (we want to give credit to the guy, whom I don't personally know). You should follow him on Twitter; he genuinely presents interesting analysis of the stocks he owns. He knows the business, and for having followed him for years, he really seems to be there for the right reasons. The fact that he also promotes cautiousness and protection of gains makes him, from our point of view, a very positive force in the fintwit world for retail investors. We need more guys like him, and bonus feature: he is making great techno mixes!

We backtested many different EMA scenarios and found that the issue with that strategy is that, in some market conditions, it triggers too much for our taste and also has, in the absolute, a low accuracy. Therefore, during those times, it was very hard to trust the signals and could easily grind a lot of capital with many successive losing trades. 2021 was one of these markets with a lot of sideways on some indices where the strategy didn't perform well. Backtesting the strategy proved that it was also a losing trade. That is something Puru Saxena never hid in his posts about the strategy, saying (and I'm quoting him), "Depending on the index, even though the upside is reduced by 2-3% per year during powerful bull-markets, by hedging, one is able to stay fully invested in his/her stocks and yet avoid nasty drawdowns. So, this may be more suitable for growth stock investors who target..." I think he is fully right on why hedging is important for growth stock investors; it's not just about the gain, but also the psychological aspect of feeling protected when the market crashes. If you've never experienced it, it's almost fun to see your portfolio doing nothing or being in the green while the market is flushing like crazy. It's usually much easier to make great purchases in the abyss than when your portfolio is decimated. Believe me, I have tried both scenarios! But still, what if we can do better than simply looking at EMA crossing points?

One of the initial versions of the hedge strategy was combining the phase angle (that I already presented) with an EMA ribbons strategy that is somewhat related to the previous strategy of EMA crossing but with a bit more information. The rationale to combine these two was that the strengths and weaknesses of these two indicators are actually complementary. Phase angle doesn't suffer from the level of the EMA that can delay the hedge signal or trigger a lot of false hedge signals. An EMA ribbon can provide a hedge signal in case the phase angle would not trigger due to a very slow but constant drawdown. We have been able to build with this a strategy that was giving a positive return on a long duration, and it allowed us to make crazy profits in 2021. The issue we had with this strategy is that it was still triggering too often for what we envisioned. Also, although this strategy was positive and was outperforming the previous strategy by a lot, the return was not that impressive considering all the trading events it was generating. I think this is normal; the indicators used in this strategy were all derived from the price action. I have played with curves for long enough, trying any math, stats analysis, etc., to know for sure that price action alone is not sufficient to accurately forecast where the market is going. As we all know, there are so many other things that influence the market, such as economics and political news, trends, governmental policy, investor financial health, etc. These external factors tend to be the ones that trigger a stock market correction but also define its floor.

So, beside the price action, we decided to start digging into alternative data that provides a more detailed and rich picture of what is currently under the hood of the stock market. If I quote Quandl, a company acquired by Nasdaq itself and that is the leader of this new trend: "We believe that data is one of the most important resources of the 21st century, and that alternative data in particular, is going to become the primary driver of active investment performance over the next decade."

What is exactly alternative data and why is it so important? An example of it and how it can provide big insights is looking at the amount of transactions on the Dark Pool. For people who are not familiar with the Dark Pool, here is a very simple and brief explanation. Some of the big transactions of public shares are sometimes done outside of the market. Often, these are transactions that would have too much of an impact on the price of the stock if done through the public market. There are other reasons to execute a transaction on this parallel market, but what you need to understand is that the Dark Pool is mostly driven by experienced, smart, and very wealthy investors. Quandl gives real-time access to what's going on in this market, and a statistical analysis of the volume of the Dark Pool can often clearly raise flags that lead the public market by one or two precious days. On February 20th, 2020, we had a three-standard deviation move on the volume (and a never-seen record on February 21st). If you look at the price action of February 20th, it was a very minor red day coming from the all-time high. With all the work that we did on the price action, I have never seen any way to trigger a hedge on this specific drawdown before February 24th.

The hedge strategy combines multiple indicators to capture a more detailed and rich picture of the market, respecting each one's strengths and weaknesses so that they become complementary. In addition, a "bear mode" has been developed that modifies or ignores certain indicators when entering a strong downtrend. This is because the market behaves differently during strong moves down, and thresholds that typically signal a new uptrend in a bull market can falsely trigger unhedge events during a dead cat bounce in a bear market. A dedicated chapter will cover this powerful indicator in detail.

Results

The combination of all the alternative data allows the hedge strategy to react quickly and confidently to market moves that may transform into massive drawdowns while ignoring smaller bumps along the road of investing. As of April 26th 2023, over the last 20 years it has triggered 40 times and has never failed to quickly capture a real market crash. The success rate in terms of positive trades is 72.5% (75% if we just go out of the market instead of going net short). While this may not seem impressive, the ratio of the return of winning trades to losing ones is more than 8.3 to 1. This means that most losing trades are very close to breakeven, providing peace of mind during smaller bumps, while capturing all the massive moves. Below are some of the data of the strategy. We chose the 2006-2010, 2016-2019, and 2019 to 2023 periods because they all encountered major corrections. In other periods with less drama, the strategy usually performs just slightly higher (around 5-20% higher) than buy and hold, which is normal since there is no real downtrend to protect from.

In addition to sending alerts when it prints a hedging signal, the strategy also alerts when entering and exiting a bear market. It also treats hedging signals during a bear market as a distinct event from a regular hedging signal. This is because a signal within a bear market carries a different level of risk than in a bull market, and should be treated accordingly. Here is the different signal that our strategy sent during the 2018 correction and Covid crash.

How WealthUmbrella use the signal ?

Like we said above, one part of hedging is having a backtested system that we methodically follow that tells us when it’s better to be on the sideline and when it’s better to be in the market. The other part is how we use this signal to actually protect ourselves. This part for us will vary greatly depending on the current market conditions. Sometimes we may be cocky and go net short, other times we will simply make our portfolio neutral. Sometimes we will use 3X inverse ETFs, other times we may short a stock. It depends on so many factors that it would be impossible to summarize all this here. So not only will we tell you when we get our signal, but we will also notify you in real-time about what we ended up doing and why we did it.

Design methodology

It’s also important to address the confidence in the strategy related to how it was designed and tested. If you were already a WealthUmbrella Bitcoin subscriber, you probably have already heard that song since we used a similar design methodology for both. As a university professor that works in machine learning, I always have a hard time trusting a predictive/reactive algorithm that I know nothing about how it was designed. The main reason for this is related to one of the main issues with machine learning/artificial intelligence/complex algorithms. For those not familiar with that field, machine learning is a way to train a network/algorithm on past data in order to be able to recognize/predict or react to future events. The best example is Apple's Siri on your phone, which learned to recognize the relationship between phonemes and words to eventually recognize what you are saying. The learning of the past data (in our case = previous drawdown) is usually not that hard. You can usually reach close to 100% accuracy on your learning data. But we actually don’t really care about how good we are at recognizing the past. The real reason we build these systems is to perform well in the future. For that reason, we always divide the data we have (for example, the SPX data of the past 25 years) into three buckets: one that we will use for training/designing the algorithm, one that we will use the data to optimize the performance during validation, and one last untouched bucket that we will just use for testing at the end. Why is it so important? As an example, I have an ongoing Master Student that worked on recognizing notches in wood panels to assess the quality of the wood. I'm skipping the detail, but he benchmarked six different algorithms. The one that was the best on the first two buckets of data had a success rate of 95% accuracy at spotting if there were defects in the wood or not. When trying the algorithm on untouched data, this went down to only 52%, which is barely better than flipping a coin. Another algorithm had a success rate of 78% on the first two datasets but was able to maintain a 76% accuracy on the never-seen data of the third bucket. That's what we actually want, since only the future is important. Going back to our problem, if you are not aware of this issue, it's easy to use all the SPX existing data to build your algorithm and think that the success you have will be repeatable in the future. The dumbest strategy (and machine learning is capable of that) would be to actually learn/memorize the date of any drawdown and when they end. You would get outstanding returns, but that would just not be repeatable in the future. When we worked on the final Hedge strategy, we initially focused only on the 2003-2009 period. Only when we were happy with the results, we then tested it back on a longer period. Applying what we did on 2003-2009 on 2009-2022 gave without any optimization instantly 90-95% of the results that you saw in the stats above. The 5-10% extra came after tuning a couple of indicators that didn't exist before 2011. By moving from SPY to SPX, which has full daily candle data down to 1962, we have also been able to validate some components of the strategy, like the bear market indicator. In summary, I know that this part about design methodology can be a bit harsh, but I hope that you learned two things:

1. With complex algorithms and learning, there is a thin line between memorization of the past and actual learning of the past that can be applied to the future. This is the main reason why a lot of algorithms out there don't really work when used in real-time.

2. We followed the proper methodology guidelines that maximize the chance that the past performance will be somewhat repeatable in the future. Future will definitely give us unforeseen surprises that we have never seen, so I don't expect to have the exact same performance, but I sincerely trust the algorithm enough that we have already put 100% of our capital behind its shield.

The last thing I would like to highlight is that I am incredibly confident that the Hedging strategy will capture the big moves down and unhedge in time to not miss all the new uptrends. These big moves are really the easiest to capture. The hardest situations are usually during consolidation periods during a bull market. These market conditions are the ones that trigger a lot of drawdowns that start brutally, thus triggering the hedge, but then find their floor pretty quickly. This can sometimes lead to going back into the market with 1-3% losses. Fortunately, at the moment of writing this text, we are still in the middle of a bear market where we can expect significant moves that are easier to see coming.