Quantitative investing in fixed income markets has slowly but surely gained traction in recent years. We talked with Olaf Penninga, portfolio manager, about the early days of quant fixed income, the challenges it brought and some of the most recent developments in this field.
“I started at Robeco as a quantitative researcher in 1998. At the time, we were looking for systematic ways to predict government bond moves. We found that there are a few fundamental drivers of government bond markets, such as economic growth or inflation, and that you need to capture investor expectations for these drivers. We also found out that expectations could best be gleaned using information from other financial markets.”
“For example, you could get information about economic growth from the stock market and information about inflation from oil prices. Data availability was more of an issue back then. We wanted to use MSCI total return equity indices, but these were not yet available on the first business day of the month. So, we had to wait until the second business day to run our model. Nowadays, of course, the data is available much faster.”
“No, the rationale was not to cut costs. We wanted to create a tool that could help fundamental fixed income investors with their investment decisions. But over time, we discovered that the best signals were often those that fundamental investors were the least inclined to consider, either because they found them counterintuitive or too difficult to reconcile with their views.”
“I remember when US 10-year Treasury bond yields fell to 4% for the first time, back in the early 2000s. At the time, this felt like an incredibly low level. But that did not keep the model from turning positive on US bonds. So, when I told portfolio managers that according to the model, they had to buy US bonds, they sent me back and asked me to redo the calculations.”
“Today, we all know that they could move below 4%. But because many people in the market shared the same belief at the time, they all got caught out. You can make a lot of money by being the first to anticipate such a move. We have seen this kind of situation time and again. Do you remember when the German 10-year yield approached 1% back in 2014? Many investors took it as a given that yields wouldn’t drop below 1%. Well we’re now in negative territory.”
“Ultimately, we found it is difficult to integrate quantitative signals in a fundamental strategy. This explains why, at the start of 1998, we decided to use the model as a standalone strategy. And it really helped. A key reason is that the strategy forces us to stick to the proven drivers of markets, especially when people are inclined to deviate from them. This has become much more common over time and is even completely mainstream now in equities. But back then, it was really exotic.”
“An interesting thing about this topic is that the researchers who investigated factors in equities in the early 1970s also looked at government bonds. And they found evidence of similar effects in the latter. For instance, the first papers describing a low risk effect in government bonds also date back to the 1970s. Only the research wasn’t taken much further. Factor effects were known about, but it has taken longer for them to be translated into factor strategies.”
“For credits, I think it definitely has to do with the availability of data. It has also taken longer for European credit markets to develop. For government bonds, I think it relates to the fact that people used to look at this market more from a top-down perspective. When, in fact, we know that the same well-known factors we exploit in equities like value, momentum and low risk, also apply for government bonds.”
We test everything we do critically
“First, we stick to our investment philosophy. We look for factors with a clear economic rationale to understand why they work. We find that factors like value and momentum can be found in all asset classes, over long periods of time, and have similar explanations. This not surprising as the same behavioral causes underlie these effects in equities, credits, government bonds, currency markets, and so on. In addition, we continue to look for long datasets to see whether we can disprove our findings. We test everything we do critically. All of which makes us confident that the factors we exploit are real findings and not based on some accidental blips in overfitted datasets.”
“We are currently focusing on factor strategies for government bonds. The groundwork has already been done and we are now looking at how to build multi-factor portfolios for government bonds efficiently and how to combine them with other multi-factor fixed income portfolios. We are also developing factor-based alternatives to passive fixed income strategies, in the same fashion as our Enhanced Indexing products for equities.”
“Meanwhile, we continue testing the performance of our duration model using increasingly longer historical data series. The idea is to prove that the model works not only in times of falling yields, as has been the case in recent decades, but also when yields rise. This way, we can ensure that the outcome of our simulations is not the result of data mining.”
“We carried out similar research a couple of years ago, looking back to the 1950s but using data for US markets only. This time we’ve expanded our scope to the global level, again with a sample that goes back to the 1950s. The results confirm the model definitely adds value across geographies, also during prolonged periods of rising yields.”
“Well, here is an example. Although the duration model has been live for over two decades, we continue to refine it. It may sound incredible, but even after 20 years, there are still questions to answer. And part of the reason is that bond markets keep changing. For one, the advent of passive investing in fixed income has given rise to some predictable patterns such as trending behavior; but also short-term reversals, just like in other asset classes.”
“Trend reversals are something we want to avoid when we implement the trend signal of the duration model, so we’ve adjusted it accordingly. The pattern in data is not clear and reliable enough to make it a standalone variable, but we at least make sure that our trend variable avoids reversals. This illustrates that as market evolve, we need to keep adjusting our models to these changes.”
This article was originally published in our Quant Quarterly magazine.
BY CLICKING ON “I AGREE”, I DECLARE I AM A WHOLESALE CLIENT AS DEFINED IN THE CORPORATIONS ACT 2001.
What is a Wholesale Client?
A person or entity is a “wholesale client” if they satisfy the requirements of section 761G of the Corporations Act.
This commonly includes a person or entity: