The Impact of Token Dilution on DeFi Prices: A Quantitative Analysis

Introduction: Why Study Token Dilution?

It is a common belief that dilution negatively affects the price of a token. This makes intuitive sense to anyone who has studied the crypto markets. Yet, while the topic is regularly debated and discussed in the crypto community, there's little quantitative study to back it up.

The question we should be asking is, why not? After all, the implications of token dilution on price are not just theoretical. They have direct and significant implications for both investors and projects. This knowledge is crucial for making informed decisions about fundraising, creating a token vesting schedule, scheduling dilution events, and more–all fundamental building blocks for shaping the future of DeFi and managing crypto investment portfolios.

In fact, token dilution may be one of the most important problems facing the crypto community. If we cannot quantify the effect of increasing our token supply, we cannot know how much the price of our token will move in response. Therefore, decisions on how many tokens to issue without negatively affecting the value of the entire company are based entirely on guesswork!

For such a crucial decision, why are many founders and investors still relying on instinct and assumptions?

The short answer to this question is that there is a serious lack of existing studies that quantify dilution's impact on token prices. The math simply hasn't been done.

The longer answer, as we discovered, was that the relationship between token supply and price is nuanced and complex, requiring multiple levels of analysis.

Our aim at Exchequer Finance is to go beyond instinct and guesses and back our tokenomics decisions with sound quantitative analysis. Therefore, we embarked on a series of quantitative studies to fill in the gaps in our knowledge. Our goal: to start at the very beginning and either validate or debunk the belief that token dilution impacts price.

In this series of posts, we're going to walk you through our process as we tried to find an accurate and reliable way to quantify the effect of dilution on token price. There is an appendix at the end of the post for those who want to get more deeply into the math. Let's get started.

Effects of Dilution on Token Price: Study Setup

Our first step in a quantitative study is to define the data to be analyzed. We focused our study on DeFi tokens because DeFi is the most visible, well-defined, and important category in crypto.

To obtain our raw data on token supply and price, we used the Coingecko API to download daily prices and market capitalization values for the DeFi category.

After excluding stablecoins and a few other miscategorized tokens, we cleaned the data by addressing gaps, duplicates, and missing values. We converted the price and capitalization value time series from daily to weekly, taking the last available value for each week and calculating the circulating supply time series for each token.

The graph shows the number of distinct tokens in the DeFi category that have existed every week since early 2015. As we can see, there has been a massive increase in the number of unique tokens since 2020. To ensure diversity, we focused on data starting on January 1, 2019.

Effects of Dilution on Token Price: Analysis

Our first attempt to define the relationship between token price and supply used a simple linear model. In other words, we assumed that for every 1% change in circulating supply, the price would immediately change X%. Simple linear regression allows us to test this relationship.

Initially, we analyzed individual token time series by regressing price returns on supply returns. Unfortunately, this approach was unworkable due to inconsistent results—some tokens showed a positive relationship, some negative, and the sign and magnitude of the impact differed depending on which period we picked. The results we got were certainly not consistent enough to draw any reliable conclusions!

Next, to see if the relationship was stable on average, we pooled the data, regressing the returns ofall token prices on all circulating supply returns for the entire period from Jan 2018 to May 2024.

Of course, this model does not explain any of the price variance (R^2 = 0). This was expected, as the primary variable affecting the price is the overall market, which we didn't include as an independent variable.

However, in the broad, average view, we can see a consistent negative regression slope of -0.4270. Not only is it a negative slope, which is consistent with our initial assumptions, but it is highly significant (P-value = 0.0012). (For those less familiar with the mathematical terms, in this case, the regression slope measures how much the price of the token moves in response to the changes in its circulating supply, while the P-value measures how confident we are about our estimate of the regression slope.)

What does this mean? The data seems to suggest that while the effect of dilution on the price of a token is overall negative, it is not a 100% one-to-one correlation. For every 10% dilution, there is a 4.27% price drop. This adds mathematically significant evidence to support our intuition that an increase in token supply causes a decrease in price.

However, after digging deeper, we once again found that this was not always the case.

Here are the results we got when we ran the same analysis on the data from a smaller subset, May 2023 to May 2024.

This particular period gives us an almost perfect negative slope of -1! We could not have asked for a better result, but it does not square with the -0.427 we got with the full set.

Let's look at a different period: May 2022 to May 2023.

This starts to muddy the waters, as, during this period, token prices actually went up when supply increased. The positive 0.5987 slope clearly goes against all our intuitions. What is happening here?

To unravel the mystery, we ran rolling overlapping regressions. Starting from January 2019, we used a one-year window for each regression, advancing by one month at a time. This meant that each dataset contained a full year of data, allowing us to capture the evolving relationship between supply and price over time. By doing so, we could observe how the regression slope and its P-value, changed month by month.

This method helped us identify trends and variations that a single regression might miss, offering a more nuanced understanding of the dynamic relationship between token supply and token price.

The chart below combines two graphs. The red and green bars represent the evolution of the regression slope. Each bar represents the regression slope for the 12-month period ending on the bar's date. If the p-value for the regression is below 5%, there is less than a 5% probability that the relationship observed in the data happened due to random variation, implying that the relationship is likely real and meaningful—therefore, these bars are colored green. For less certain results, the bars are colored red.

As the left graph axis shows, the regression slope is mostly negative and often statistically significant.

To better interpret the regression slope's behavior, we overlaid its graph with the price of ETH, which is shown in the log scale as a blue line.

While the regression slope shows great variation and often flips signs, it tends toward -1.0, which is what our intuition tells us it should do. If we increase a token's circulating supply by 10%, its price should drop by 10%.

This behavior tracks with what we saw in the initial phases of our analysis. For most periods, the effects of dilution followed the pattern our intuition leads us to expect: as token supply goes up, prices go down.

However, there are also periods when the market behaves completely differently than expected. As we can see on the graph, during the crypto winter, the pattern completely broke down. Between July 2022—a data set that, because of the 12-month rolling regression, included data as far out as July 2021 and Dec 2023—the slope was much smaller than -1 and even became positive.

What can we learn from these results? Does this mean that during the crypto winter, token dilution had no effect on price? It is unlikely. Rather, it's clear that the relationship between token supply and price is nuanced and complex, and we need to take a different approach to measuring it during this unusual period.

Of course, we also need to try to understand why the crypto market behaves this way. And we do have a few ideas.

Effects of Dilution on Token Price: Our Hypotheses

We propose two explanations for why simple regression might not capture the relationship between increasing circulating supply and token price.

First, during the crypto winter, the severe market downturn disrupted the expected relationship between token supply and price. As token prices dropped by 90% and all tokens moved down in unison, they became desensitized to current supply changes because there was no more room to react—with the entire market down so much, an additional reaction to an increased circulating supply is hard to detect.

However, this does not necessarily mean that projects with already bloated circulating supplies weren't affected even more; it just means the linear regression model failed to capture the relationship accurately.

Second, when we consider individual tokens, not just ETH or BTC, an increase in circulating supply is, in general, asynchronous with a change in token price. For example, prices might lead if a token incentive program is announced in advance or closely watched, reflecting anticipated supply increases, or prices might lag if a gradual increase in token supply goes unnoticed at first by the community. On top of this, the large overall number of tokens and trading venues makes it harder for individuals and organizations to track each protocol, leading to market fragmentation. This doesn't mean that increased supply won't eventually affect token price, but unlike in TradFi markets, it may take time for the effect to manifest.

In the coming posts, we'll leverage non-linear methods that account for these nuances to determine whether our hypotheses are true.

Conclusion

The first part of our study brought up a host of questions, but we also came to several important conclusions.

For those less interested in math, the most important thing to note is that our initial linear regression analysis does align with our intuition: we found reliable evidence that, in general, increasing token supply drives prices down. However, we also found just as reliable evidence for periods when the market performs unexpectedly.

This means that if you are relying on simple intuition to make decisions on investment and token vesting schedules, you are likely to be pointed in the right direction most of the time–until, when a more localized non-linear model is required, you're not.

For the mathematically inclined, it became clear to us that the simplest model often falls short of being able to explain reality. While the linear regression approach provided limited but undeniable success, it also pointed out the complexities that a simple model can't capture. Therefore, our next steps for this study involve exploring alternative methods to better understand and quantify the effects of dilution on token supply.

In the next posts in this series, we'll share those methods and explore models that account for dilution's asynchronous effects and differences in its impact in bear and bull markets.

Finally, zooming out, the lack of studies that quantify the effects of token dilution highlights a need for a more thorough investigation. At Exchequer Finance, our aim is to bring that clarity to the community. If studies and analysis don't exist, we will do the work ourselves. If you're interested in keeping up-to-date on our research to make crypto investing and founding more quantifiable, subscribe to our blog and follow us at www.exchequer.fi.

Appendix

Before running regressions on the data, we wanted to ensure their distribution was reasonably close to normal. To achieve this, we filtered outliers by dropping extreme values of the entire set:

def remove_outliers(df, column, multiplier=1.5):
    Q1 = df[column].quantile(0.25)
    Q3 = df[column].quantile(0.75)
    IQR = Q3 - Q1
    df_out = df[~((df[column] < (Q1 - multiplier * IQR)) | (df[column] > (Q3 + multiplier * IQR)))]
    return df_out

The figures below show histograms of price and supply returns from January 1, 2019 onwards. The graphs clearly show that the outliers have been removed.

A quick check of Q-Q plots confirms that while the data is not normal, it is not unreasonably so.