The Gateway to Algorithmic and Automated Trading

Regime changes in automated trading

Published in Automated Trader Magazine Issue 42 Q1 2017

The term structure of interest rates provides many opportunities for systematic traders. The level, slope, curvature and volatility of interest rate markets are all heavily regime-dependent. Early identification of changes in regime is key for developing successful trading strategies.


Joseph Choi

Joseph Choi publishes the Curve Advisor strategic newsletter. He previously worked as a senior proprietary trader in J.P. Morgan's Global Currencies and Commodities Group. For years he traded well over 10 million Eurodollar futures and options contracts annually - by hand, without an algo. Mr. Choi is still recuperating.

Automated trading is backward-looking. It is typically based on historical data - whether it is pricing data, behavioral data or both. It is presumed that by understanding the secrets of the past, one can succeed in the future. Depending on the algorithm and the markets, this could be somewhat true. However, there are important considerations that should be taken when analyzing interest rate historicals. This is an introductory discussion of the importance of understanding interest rate regimes when looking at past data and trying to apply it in the future.

For the purposes of this article, I will refer to the US Federal Reserve and Eurodollar interest rate futures. But the concepts can apply to other central banks and other interest rate products.


Many traders tend to think mostly of one dimension - up or down. But there are other meaningful dimensions with interest rate trading because of the availability of: (1) various settlement dates (for example, there are liquid short-term interest rate futures in quarterly intervals out to five or more years), (2) options for all strikes for each of those contracts and (3) spreads between various interest rate instruments (across products and countries).

One of the beautiful features of interest rates is that there is a strong relationship between one part of the yield curve and an adjoining part of the curve. The fact that there are so many interrelated contracts to choose from gives us an opportunity to find additional value. Consider the following betas and correlations for ED13 (the 13th generic Eurodollar futures contract) for the past year in Table 01. If we had to hedge a position in ED13, we have a number of reasonable alternatives - especially in a high-frequency trading setting. One also has many more opportunities to scalp a tic if an eager counterparty were to trade a contract on the wrong side of the bid/ask spread in a particular contract.

Table 01: Beta and correlation of various Eurodollar futures contracts to ED13 (10 February 2016 to 9 February 2017)

Table 01: Beta and correlation of various Eurodollar futures contracts to ED13 (10 February 2016 to 9 February 2017)

This interrelationship gives us an opportunity for high expected value positioning when the relationship between various contracts are at attractive levels. Analyzing the slopes and curvatures of the various parts of the yield curve can be used to find value or to hedge more profitably. Consider, in Figure 01, the equally-weighted ED17-ED21-ED25 one-year butterfly (buying ED17, selling two times ED21 and buying ED25) at three basis points, as was the case earlier this year (red dot). If you were short ED21, trying to convert to the butterfly structure (by buying 50% of ED17 and 50% ED25) would have a positive expected value in the current rate environment. Historically, we have seen many more observations above the red dot than below it. There are many ways to get a similar exposure, using various combinations of the contracts between ED17 and ED25 (three-month flies, six-month flies, nine-month flies, weighted flies and weighted flies with uneven legs). There can be many attractive structures to trade on the yield curve at a given point in time.

Figure 01: ED17-ED21-ED25 fly vs. ED2-ED10 spread (5 yrs)

Figure 01: ED17-ED21-ED25 fly vs. ED2-ED10 spread (5 yrs)

Note that in Figure 01, the trading range of this structure over the past year (blue triangles) is markedly different than from the past five years (grey diamonds). This is because we are comparing different interest rate trading regimes. The past year has been an environment where the Fed hiking cycle had begun, while the preceding four years had generally been an environment where rate hikes were not expected in the near-term. We also had some structural shifts on the yield curve, that I will discuss later in the article.

Does the red dot seem like an attractive trade? Yes. But only if the current interest rate regime is similar to that of the past five years. It is possible that the rate regimes are different. If we look at this structure over a larger period, you will see there are periods where the structure has been noticeably negative. Figure 02 shows this same structure over the past 17 years. The red dot could still be considered somewhat attractive, but not as attractive as before.

Figure 02: ED17-ED21-ED25 fly vs. ED2-ED10 spread (17 yrs)

Figure 02: ED17-ED21-ED25 fly vs. ED2-ED10 spread (17 yrs)

When you take too large a data sample (say the 17-year sample), you can miss many high probability opportunities from taking too wide a view. Conversely, when you become overly dependent on too small a data sample (say the one-year sample), you can severely underestimate the risk of a trade. So how do we know what historical data sample is appropriate?


The nature of the relationship between contracts can change depending on the interest rate trading regime (hiking/easing/neutral environment) we are in. The possible shape of the yield curve is a function of various paths of future Federal Reserve policy. Figure 03 shows the history of Fed hiking and easing cycles. You can see a cyclical pattern of rates rising and falling, as the growth and inflation conditions of various business cycles have dictated. As interest rates move from lower to neutral to higher to neutral to back down lower, the relationship between the parts of the yield curve will change.

Figure 03: Federal funds target rate

Figure 03: Federal funds target rate

Just looking at the overall historical data can be like mixing your steak, potatoes and vegetables in a blender before consuming. A large heterogeneous data sample will contain a mix of various regimes, many of which may not be relevant to the current environment. While looking at all regimes simultaneously may be useful, you also may lose opportunities from not understanding the flavor of the potential curve moves in a particular rate regime.

In Figure 01, we saw that the blue triangles and grey diamonds give dramatically different pictures of the ED17 one-year fly. If you use any models aggregating historical data, the outputs can also be misleading. As a simple example, you can see that the slopes of the triangles and the diamonds are very different, and hence the regression betas can vary noticeably depending on the sample period. A more accurate picture of yield curve historicals comes from understanding the importance of the various interest rate regimes when looking at the data.


The largest driver of the shape of the yield curve is whether the Fed is expected to ease, hike or maintain its policy rate at a given point in time. This major regime is straightforward to identify, as the Federal Open Market Committee (FOMC) will be looking to hike, ease or leave rates unchanged. The FOMC shows their future intentions via their statement and dot plots (a recent innovation). When we get a major regime change (an example would be when we go from a neutral rate environment to a hiking environment), the relationship between the various parts of the yield curve can change noticeably. A major regime change occurs every few years. So within that time range, we can more profitably exploit interest rate curve moves by understanding what the constraints of that regime imply for the curve.

It is possible to switch regimes very quickly. From a hiking regime, we can switch suddenly to an easing regime on a catastrophic event. It may be possible to find cheap protection for such events so crisis (outlier) protection should always be considered. It is much less likely to go abruptly from an easing cycle to a hiking cycle. The Fed generally needs at least several months of data to start changing gears.

As an example of how regime changes can affect curve structures, consider the graph of the ED9-ED13-ED17 one- year butterfly in Figure 04. The green shaded areas represent hiking cycles and the red shaded areas represent easing cycles. You can see that in hiking cycles, the butterfly tends to get very low and hovers around zero. In easing cycles, this butterfly tends to be at the higher end of the range.

Figure 04: ED9-ED13-ED17 fly with indication of major (coloured regions) and minor (arrows) regime changes

Figure 04: ED9-ED13-ED17 fly with indication of major (coloured regions) and minor (arrows) regime changes

These observations make sense from a logical perspective. Even if you did not see this chart, you could expect the curve to move this way from considering the future path of Fed rate hikes. When the Fed is hiking, most of the hikes will be priced into the very front end of the curve (first eight contracts, or two years), and the back end of the curve (the subsequent twelve contracts) will be much flatter. If it looks like the Fed could overshoot on hikes, you should be able to picture how this structure could go negative, even though we do not see it on the graph. An overshoot in rates could be followed eventually by eases further out, and could depress curvature on this part of the curve. As always, just because it's not in this small sample of historicals, it doesn't mean it can't happen. As the Fed is easing, eventual hikes will be priced in further out the curve, resulting in higher curvature. An exception could be if things looked so dire that hikes were priced out for an extended period of time, similar to 2012 when the year flies went noticeably negative.

The periods between hiking and easing cycles typically start drifting towards the next cycle. From the end of an easing cycle, the butterfly will eventually drift lower towards a hiking cycle. From the end of a hiking cycle, the butterfly will eventually drift higher towards an easing cycle. The most notable exception would be in the middle of 2013, where the butterfly spiked back higher, despite moving towards a hiking cycle. This was Bernanke's 'taper tantrum' where yields on the long end exploded higher. If the start or end of a cycle is unclear, or if the nature of the next cycle is unclear, you may see curves resembling a probability-weighted result between the scenarios.


Within each major regime, there can be several minor regime changes. A minor regime change can be anything that causes a major repricing of a part of the curve. Some examples include:

1) Major central bank interventions

These could be quantitative easing (QE), tapering or altering reinvestments. When you look at the three US QEs in late November 2008, November 2010 and September 2012, one can see noticeable dips in the butterfly curve (black down arrows in Figure 04). Buying on the longer end of the curve will generally result in less curvature. Note from the price action that the markets started anticipating the later QEs and the structure went higher soon thereafter. You can also see a dip around January 2015, which is when the European Central Bank (ECB) started its QE program (blue down arrow). The markets are becoming more global and because of influences like interest rate parity, large central bank actions in other countries will also affect the US curve. We would expect a dip on any further major central bank QE stimulus. Conversely, we would expect a bump higher whenever longer end stimulus is removed. June 2013 was the taper tantrum, and we see the large spike higher (purple up arrow). Other examples of longer end stimulus being removed include the Fed ending reinvestments of its balance sheet, the ECB starting tapering and the Bank of Japan starting tapering. The Fed and ECB may act as early as later this year.

2) A Fed that seems intent on hiking gradually

All other things being equal, a central bank that philosophically chooses to proceed cautiously pushes hikes further out the curve. Fed Chair Janet Yellen has been one of the more dovish members of the FOMC for quite some time. We could see this "gradual" regime change when Yellen's term ends early next year and President Donald Trump selects a more hawkish Fed Chair.

3) Electing a new stimulus-minded President

Something I told clients on the morning of the US presidential election results was that we are in a regime change and the past (historical) ranges on trades may not hold. We are still in a hiking environment, but the future path of Fed hikes (and government spending) has changed dramatically, with the specter of large future tax cuts, deregulation and large stimulus spending. The election upset the typical flattening in the longer end of the curve you would see in a hiking cycle (yellow up arrow).

As you can see, what originally seemed like a lot of noise in the chart of this butterfly looks more reasonable after factoring in the minor regime changes. This is all information that could be used more effectively in interest rate modeling. There may be other types of minor regime changes. As Justice Stewart once famously said, "I know it when I see it". The minor regime changes discussed do not occur very frequently and many of them can be anticipated in advance. If a particular trading regime does not seem stable, one can always reduce risk. There is no rule that says an automated trading program needs to be running at the same risk at all times. Just knowing the larger factors affecting the current curve regime could help you become one of the first in the markets to recognize when those drivers have changed. A better understanding of the drivers of the yield curve could be leveraged to manage risk more effectively and could provide better trading opportunities.


In addition to looking at major and minor regime changes, there are some structural changes that have affected the interest rate curve in all regimes.

1) Neutral rate

Five years ago, the FOMC's longer run Fed Funds projection (the Fed's long term equilibrium rate) was at 4.25%. It is currently 3%. Part of the decline in the neutral rate is a decline in US potential GDP. Ceteris paribus, a lower neutral rate puts downward pressure on positive butterfly structures. If we were to go back into recession, you would not expect the ED9 one- year fly to get as high as in the previous easing cycles. It would also make it virtually impossible in Figure 02 for ED2-ED10 spread to get to 400bps, as the historicals may indicate. It is not clear how quickly economists will revise their estimates of the neutral rate or if the neutral rate is just a data fitting exercise. But the neutral rate typically does not change very quickly.

2) Inflation

There are some structural reasons for lower inflation, such as better pricing discovery, globalization, technology improvements and demographic shifts. All other things being equal, you would expect the hurdle for elevated inflation to be a little higher than in the past.

3) Demographics

The first of the baby boomers turned 71 this year. The last of the baby boomers is only 53. So for some time, pension funds and insurance companies may need to buy the long end of the curve. People are also living longer. These factors will keep some bid to the long end. The demographic demand for longer term fixed income won't change in the near future.

The main factors to watch for here are the neutral rate and inflation. I am not referring to the monthly economic data prints, although they are important. What I am talking about are any reasons to think the structural forces that have put a weight to growth and inflation could change. Trying to increase productivity has been discussed as a solution to the anemic GDP. This again should not change very quickly. You could consider adjusting some past historicals to account for these structural changes in the neutral interest rate.


Value-at-Risk (VaR) and other history-based models of risk assume that the recent history captures the range of (daily) possible moves of the interest rate curve. However, any time there is a regime change, future ranges of moves can be different from the recent historical patterns. Future risk can be overstated or understated, depending on the trade and the regime we are leaving and entering. We saw in Figure 01 that the trading range for the ED17 one-year butterfly can vary tremendously depending on the sample period. Risk managers need to be especially aware when the interest rate curve is in the process of changing regimes. Calculating too small a VaR as compared to the future risk in the current environment will expose the firm to too much risk, while assigning too large a VaR will result in severe undersizing on a trade. Loss minimization is important, but in this highly competitive trading environment, profit reduction can be almost as lethal.


This article scratched the surface of the topic of regime changes. I wanted to give you some food for thought when analyzing interest rate historicals. Using the entire historical database could be useful - especially since the trading regime can sometimes change quickly. However, there will be many opportunities you miss by trying to incorporate all of the data. Looking forward from the backward-looking historicals to incorporate some easy-to-identify regime change markers can provide some clarity and valuable insight into opportunity. As we have seen from one example, just factoring in some basic regime adjustments can help explain away some of the noise in the data. Understand the current regime we are in relative to the historical data and understand what regimes we are likely to progress towards.