🔍Deep Dive: Bitcoin's Funding Rate

Welcome to this week’s edition of the deep dive where we conduct a statistical analysis of some of the most relevant metrics in the Bitcoin ecosystem.

This week, we explain how to use data from the futures market to better understand Bitcoin's market structure and sentiment dynamics.

The futures market is important to understand because it allows users to bet on Bitcoin’s future price. This data subsequently contains a robust amount of speculative data that we’re able to model out to determine whether or not the futures market directly impacts Bitcoin’s short-term spot price movements.

In essence, while the long-term movement of Bitcoin’s price is determined predominantly by “hodlers”, short-term price movements are driven by speculators. In order to make well-informed decisions, we need to understand the behavior of both types of market participants. Let’s dive in!

Table of Contents

1. Backstory

2. Futures vs Perpetual Swaps

3. Funding Rate

   1. The Formula for Funding Rate

   2. Analyzing The Funding Rate

4. Statistical Analysis

   1. Limitations and Revisions

5. Key Takeaways

⌛️ Estimated Read Time: 0.9 Blocks (~9 Mins)

Backstory

Back in Q3 ‘22, with Bitcoin’s price hovering around the $20K mark, many market participants began calling the bottom in Bitcoin’s price. Their reasoning was based on previous historical analysis pointing out that Bitcoin had never dropped below its previous cycle top of ~$19,800. Thus, invalidating this trend was incredibly unlikely.

Sure enough, we experienced a prolonged dip below the previous cycle top, catching many off guard as the majority of the world left Bitcoin for dead (yet again). Amid all the chaos, however, there was one metric I was monitoring that offered insight into how much further downside was likely left.

This metric was the funding rate in Bitcoin’s perpetual swap market, which had dipped to unbelievably low values. The 3-day moving average suddenly crashed by an annualized 40%, signifying that sell-side pressure was very likely exhausted (green box below).

I used this metric (amongst others) to significantly ramp up my Dollar Cost Averaging (“DCA’ing”) into Bitcoin back in Q4 ‘22, and it continued to be helpful in understanding market moves. The funding rate is, therefore, a metric I personally monitor very closely. But what exactly is this rate, and how is it measured?

Futures vs Perpetual Swaps

Before defining the funding rate, we need to understand both the futures market and the perpetual swaps market (don’t worry if you’ve never heard of these terms before; they’re relatively simple to understand).

In layman’s terms, the futures market is a marketplace where individuals can make bets on the future prices of a security or commodity using what are known as “futures contracts”. Typical futures trades involve participants entering into a contract where the owner of a commodity (e.g., a wheat producer) agrees to physically deliver a certain amount of commodity on a certain date and at a certain price. This gives both sides certainty about the future.

However, in many instances, a trader only wants to speculate on the price of the underlying asset, and the fact that the contract involves receiving physical delivery poses an unnecessary challenge.

This is why “perpetual swap markets” developed, giving market participants the ability to speculate on the price of assets (including Bitcoin) without ever having to actually physically own it. Unlike traditional futures contracts, perpetual swaps do not have an expiration date, which means traders can hold their positions indefinitely. These contracts are designed to trade close to the spot price of Bitcoin by employing what’s known as a “funding rate” mechanism.

Funding Rate

The funding rate (“FR”) is a periodic payment made to either long or short traders. It is calculated based on the difference between the perpetual contract price and the spot price. The primary purpose of the funding rate is to ensure that the price of the perpetual swap stays close to the underlying asset's spot price.

The Formula for Funding Rate

Each exchange calculates the funding rate a little differently, but here is how Bitmex computes it:

First, they begin with defining the Interest Rate (I) based on the below

I = (Q - B) / T

Then they define the Funding Rate as:

F = P + Clamp (I - P, -0.05%, 0.05%)

Where

• F = Funding Rate

• P = Premium Rate (adjusted by I up to 0.05%)

The funding payments occur at regular intervals (typically every 8 hours), and the direction of the payment depends on whether the funding rate is positive or negative:

• Positive Funding Rate: Long traders pay short traders.

• Negative Funding Rate: Short traders pay long traders.

Analyzing The Funding Rate

The funding rate provides valuable insights into market structure and positioning (i.e. is the market ‘offside’ and if so, can we benefit from this potential mispricing).

To answer this, we gauge sentiment, leverage and arbitrage:

1. Market Sentiment:

   • A consistently positive funding rate suggests that long positions are dominant, indicating bullish sentiment.

   • A consistently negative funding rate suggests that short positions are dominant, indicating bearish sentiment.

2. Market Leverage:

   • High absolute values of the funding rate (either positive or negative) imply that there is high leverage in the market. This could indicate potential volatility due to forced liquidations if the market moves against these highly leveraged positions.

3. Arbitrage Opportunities:

   • When the funding rate is significantly different from the interest rate component, it may create arbitrage opportunities. Traders can take advantage of these differences by entering positions in the spot and perpetual markets to profit from the funding payments. This is the basis for the cash and carry trade [see link to our previous article covering this topic in more detail].

Let us now explore the relationship between Bitcoin’s price and the funding rate, by charting price against the funding rate's 3-day moving average ('“3DMA”). We use the 3-day average (pink line in the chart below) because using a single-day observation results in a chart with too much volatility. Hence, smoothing results over a longer period of time allows us to ignore statistical outliers and draw more reliable conclusions.

Based on the above chart, we observe that:

1. The funding rate is highly correlated with Bitcoin’s price. This is logical because as Bitcoin’s price moves higher, it attracts new speculators to the market who in turn take long positions that drive the FR higher.

2. Price also reacts directly to the funding rate. As FR increases rapidly to high levels this signals a potentially overcrowded market which could indicate a reversal.

Based on the above observations we may be inclined to use the FR as a contrarian indicator, however we first need to run statistical models to confirm whether or not the FR has any predictive value.

Before doing this, we can chart the data (see chart below). At first glance, the FR certainly appears relevant, with many of the local tops and bottoms coinciding with the extreme values of FR.

To enhance legibility, we’ve colored the below price chart based on funding rate values. High FR values (+20%) are represented by green circles (positive FR) and red circles (negative FR). Larger circles indicate larger FR values.

As we can see, FR has been an excellent indicator of market reversals. Similar to all short term metrics, it isn’t accurate 100% of the time, but it certainly captures important reversal points.

Statistical Analysis

Having created the charts, let’s now confirm the above using statistical models (we are data-driven at the end of the day).

The process begins with us performing a regression of Bitcoin’s price change relative to the funding rate (“FR”) where:

• Price change is represented as the percentage change in price 7 days into the future (“PriceChange”). By using the change in future price the model is able to accurately handle autoregression.

• The model’s “predictor” is the 3DMA of the annualized funding rate (“FR3”).

This regression shows a negative coefficient for FR3, which confirms our prior observation that a higher FR is typically bearish for Bitcoin’s short term price trajectory. Based on the regression model, this would imply that for every 4% increase in FR, there would be a corresponding 1% drop in price seven days later.

In statistical analysis, we’re typically aiming to achieve higher R² in order to be able to draw more reliable conclusions from the data (R² basically measures the portion of the variation in the actual data that the model was able to capture).

Unfortunately for us, the R² of this regression model (shown below) is only 0.0068 which is very low (although not surprising given that we’re using only a single variable model to predict something as complex as Bitcoin’s price).

In order to improve the R² , we can include a wide variety of metrics (e.g. lags, non-linear terms, interactions, etc.). This is a complex process and can take significant amount of time, so we’ve examined hundreds of models, cross-validated them to guard against overfitting, and triangulated the results with machine learning prediction models.

The result is that there are two variables that, when included together in a regression model, produce highly accurate predictive results. These are the 3- and 14-day moving averages of FR (“FR3” and “FR14”).

With these variables, the R² jumps up to an impressive 0.2 (see table below). This implies that with only two variables extracted from the futures market, we’re able to model 20% of the Bitcoin’s price movements seven days into the future!

The regression results also indicate that FR3 and FR14 have opposite coefficients. In other words, a higher FR14 is price positive, yet a higher FR3 is price negative.

The reasoning behind this is that initially when speculators enter the market (taking on leverage, open long positions) this is initially supportive of Bitcoin’s price. This is captured by the FR14 variable, which measures the longer-term trend of speculator behavior.

However, when speculation increases rapidly over a short time period, the FR3 captures this excess short-term speculative behavior, raising the FR above its longer-term trend. Essentially FR3 captures the spikes in speculative behavior over and above the trend.

So, when FR14 is positive and FR3 is negative, the overall trend is positive. However, toward the end of the 14-day period, the funding rate rapidly reverses and enters negative territory, signaling a likely price reversal in the next few days.

In other words, it is the divergence between FR14 and FR3 that is really driving the direction of Bitcoin’s price.

We created an updated model below which incorporates this divergence, showing only the points where the model predicted at least a 2% increase (green) or decrease (red) in price.

This chart more effectively captures many of the local tops and bottoms as shown by the green and red bubbles. One excellent example was the November 2022 crash, where it flashed red right before the crash and then proceeded to flash green at the bottom several times until January 2023.

Limitations and Revisions

As with all statistical models, this two variable FR model is not the be-all and end-all of metrics, especially with an R² of only 0.2. However, it does offer useful insight that may be worthwhile including alongside other datasets to help improve the probability of making the correct investment decision. Our goal was to only gain insight into short-term price movements and so we have not included many other variables that might explain medium- and long-term price movements.

Models should also be stress-tested and cross-validated. We spent considerable time cross-validating these results, leveraging Random Forest (a machine-learning model with an outstanding track record), to do so. This resulted in better performance than a typical econometric model, producing a test set R² of as high as 0.4!

This increases our confidence that the model can reliably predict on data it has never seen (i.e., likely to perform well on future data), and will form the basis for one of Bitcoin Insights’ proprietary data tools.

Key Takeaways

The funding rate is a key metric in understanding Bitcoin’s market structure. By understanding the FR we’re able to gain insights into:

• The extent of leverage in the market;

• Whether sentiment is skewed bullish or bearish

• Whether any arbitrage opportunities exist.

Understanding the funding rate requires an understanding of both the futures market and perpetual swaps market, with the difference being that perpetual swaps don’t require physical delivery of the asset (Bitcoin). This provides the trader the benefit of holding their long or short position indefinitely.

After compiling the data into various chart, we observed that the FR was highly correlated to Bitcoin’s price, shown clearly by the top (red signal) and bottom (green signal) indicators.

Aggregating this data into a linear regression model we utilized the FR3 and FR14 metrics to substantially improve the R² of the model to 0.2, which explains up to 20% of Bitcoin’s price movements up to 7 days in the future.

Finally, we cross validated the linear regression model further by utilizing machine learning to improve the R² even further to as high as 0.4.