Why You Should Use Range Bound Strategies

Last week we launched TokenSets, the first platform that enables traders and crypto enthusiasts to automate asset management strategies by simply acquiring an ERC20 token. Many of you have asked about how the optimal parameters were determined for Range Bound strategies. We’ve prepared the analysis below for an insight into how the tolerances were determined.

Disclaimer: The content below is provided for educational purposes only, and not indicative of future performance. None of the following should be interpreted as investment advice. The tools used below follow a predefined set of parameters and isn’t actively managed by Set Labs Inc.

TL;DR

• Important to evaluate trading strategies on risk adjusted returns. Lower variance average returns lead to higher overall returns.
• Improved Performance in neutral markets: Range Bound Sets have performed better than holding Ethereum or Bitcoin in neutral markets.
• Parameters are determined using generated market data. Monte Carlo simulations used to create many potential price “histories”. Prevents overfitting on historical data.

What is a Range Bound Strategy?

The purpose of a range bound strategy is to profit when prices of a volatile asset (i.e. Ethereum, Bitcoin) are not trending in one particular direction. Each Range Bound Set is comprised of a mix of the volatile asset and a stable asset (i.e. Dai). This strategy automates buying the bottom and selling the top within a range, capitalizing on price movements even when the underlying asset doesn’t change in price over the long-term.

In order to execute this strategy in Range Bound Sets, we define a static lower and upper bound that triggers a rebalance once the volatile asset changes in price to exceed those bounds (thus signifying being at the edge of a range). Intuitively, it makes sense how this strategy makes money if you trade in a range since you will effectively be buying low and selling high.

But how does it perform in the wild and how do we select the correct parameters for a given “theory” of market price movement?

How to Evaluate a Strategy

In order to make a decision about what bounds are best, we need to decide what our evaluation criteria is. Total returns are obviously very important but equally important, and often overlooked, is the consistency of those returns.

For example Bitcoin and Ethereum could both average 10% returns but say Bitcoin returns 5% and 15% over two months and Ethereum returns -5% and 25% over the same two months. If you calculate the total returns over that period Bitcoin returns 20.75% (1.05 x 1.15 = 1.2075), while Ethereum only returns 18.75% (0.95 x 1.25 = 1.1875), so there’s clear advantages to consistency of returns.

Given the need for top line returns and also the consistency of those returns, a typical metric used to evaluate strategies is the Sharpe Ratio. At a high level the Sharpe Ratio measures the average expected return per unit of variance of those returns. In other words, we are trying to measure the amount of returns vs. the riskiness of those returns.

A higher Sharpe Ratio indicates better risk-adjusted returns.

How Do Our Sets Fare?

In order to evaluate the effectiveness of our Sets we compare the Sharpe Ratio of the Set strategy to the Sharpe Ratio of a benchmark strategy. For our Range Bound strategy we’ll compare to holding an equivalent dollar amount of the underlying volatile asset.

To start, let’s examine a time period spanning the back half of 2016 when Ethereum bounced around but rarely trended in one direction. Comparing $100 of each asset you can see how the value changes over time for our low volatility and high volatility strategy:

As we can see both strategies are more profitable over the given time period because they can capitalize on the price movements of Ethereum. Now let’s compare the Sharpe Ratio of these two strategies. We’ll be using monthly returns to calculate the Sharpe Ratio:

Both strategies offer better risk/reward than holding Ethereum in these types of market conditions.

Now, let’s examine a time period that spans the run up and subsequent correction spanning May 2017 to present:

The returns for the Range Bound strategies are clearly better over the long term and they appear to be much less volatile as well. However, the Sharpe Ratios tell a different story:

Interestingly, it appears that on a monthly returns basis even though the strategies are more profitable and appear to be less risky, the large monthly gains in a very high volatility environment out-weigh the elevated risk. This brings up an important point about time horizon, and why these strategies are currently designed for longer holding periods. When we decide to analyze three month returns, which implies a longer holding period for token holders, the Sharpe Ratios change again:

Here, you see the Range Bound strategies start to come much more in-line as the large monthly gains are smoothed out. Using these real world examples to form an intuition, we can then dive into the methodology for selecting the tolerances for our Sets.

Determining Parameters

The two examples above illustrate that the defined parameters on the Range Bound sets are profitable under conditions where prices don’t ultimately change over the long term. However, deciding the rebalancing tolerances off of that is insufficient because the results would overfit the specific price movements of those time periods, and in fact, the selected tolerances were not the best performing tolerances for those periods.

In order to not overfit the results, we instead took the various histories outlined above and were able to imply a distribution of daily price changes. By sampling this distribution in a Monte Carlo simulation, we can approximate many potential price histories to test against for optimal parameters. Furthermore, because the range-bound strategies are designed to profit in range-bound time periods we limited the price histories to those where prices ended within -30% and 50% of the starting price.

The chart above shows the calculated Sharpe Ratio for a range of lower and upper bound tolerances. Each upper tolerance was paired with each lower tolerance to explore a good range of possible tolerances. The upper tolerance was relatively definitive, with the model suggesting to take profits when the Ethereum/Dai mix weights 57% in Ethereum’s favor (indicating a 32.5% price increase from last rebalance). The ETHHIVOL set’s upper tolerance is configured to 58% (indicating a 38% price increase from last rebalance) based on slightly more volatile data and looser bounds of the market outcomes.

The lower tolerance is much murkier, and is very sensitive to changes in how one quantifies risk, because when the lower tolerance is breached, the portfolio is adding more of the risky asset. As shown above for ETHLOVOL there is very little separating 40% and 42% Ethereum weight (indicative of a 33% and 27.5%, respectively), and in this case we opted for slightly more active management.

For the ETHHIVOL Set the lower bound is configured to 33 (or a 50% downturn), which is partly due to higher volatility and looser final price bounds, but also due to allowing users a different way to quantify risk. The ETHHIVOL Set calculates risk as the variance of the raw percent return, whereas the ETHLOVOL calculates using the log¹ of those returns. Due to the fat tail of returns, especially to the upside, the raw return values imply a much higher variance and thus push the lower tolerance further down. This Set is better for those that want to protect against the high magnitude price moves more prevalent in crypto assets.

Conclusion

There are a lot of considerations to take into account when deciding which token makes the most sense for any given user. Ultimately it is up to each user to decide how they expect prices to change, how long they want to hold assets for, and how they think of risk. As Set continues to grow we hope to add more assets to satisfy user needs for different strategies (i.e. moving average based strategies) that more closely track with each user’s view of market direction and analysis.

¹ Log is used because it adheres closer to Sharpe Ratio’s underlying assumption that returns are normally distributed, to the detriment of fully capturing the fat-tails present in returns.