TLDR

• Bitget's on-chain trading volume impacted its user growth, adding 2.5 million in Q2 2024, with over 100 tokens listed since April via platforms like PoolX and Premarket.
• Bitget's research highlights token listings criteria including market traction and leadership influence.
• Nansen's Fama-MacBeth regressions confirm Ethereum's TVL and fees as predictors of short-term token prices, linking on-chain metrics to governance token values.

Introduction

In this Research note, Bitget and Nansen Research analysts tackle two dimensions:

• the analytical inputs that an exchange considers in deciding a token’s potential (part I)
• the relationship between various on-chain and off-chain data and governance token price moves (part II)

I. Token Listing Process for Centralised Exchange

By Bitget Research

In the first part of this note, the Bitget team highlights its selection process when considering listing a token, as well as case studies for real-life applications.

Bitget Listing Criteria

Bitget’s approach to coin listings aims to deliver maximum value under controllable risks. The core principles guiding Bitget's coin listings are identifying assets with growth potential, listing popular assets quickly, and offering a comprehensive range of options to users.

• Growth Potential: Bitget prioritises listing coins with upside potential, whether through short-term price appreciation or sustained growth driven by robust fundamentals. This ensures that users have access to assets that offer potential profit opportunities.
• Timely Listing: Bitget strives to be among the first platforms to list promising assets, aiming to secure listings before other centralised exchanges. This allows traders and investors to seize new opportunities early on.
• Comprehensive Offerings: Bitget provides a diverse range of mainstream coins and tokens, catering to various trading needs. This one-stop shopping experience ensures that customers can easily access popular assets in one place.

Bitget's coin evaluation process is based on quantitative standards and incorporates both word-of-mouth recommended projects and assets identified through research efforts. The research team employs an automated on-chain data monitoring system to track asset movements in real-time, ensuring early identification of trending assets across different sectors.

Bitget’s Methodology for Token Screening

The Bitget research team has set up an automated on-chain data monitoring system, enabling a first-hand grasp of on-chain assets’ movements. It helps evaluate coins for listing in five key dimensions:

1. Market Traction: Assessing whether a token shows a robust market dynamic before or after it is issued.
2. Community: Verifying the authenticity of on-chain data by analysing community and sentiment.
3. Technological Innovation: Evaluating the innovation introduced by the project and its integration with other ecosystems to drive mutual growth.
4. Token Economics: Assessing token pricing rationality and distribution to examine the potential resilience of the project. 5 Security: Evaluating token risk factors such as contract security, liquidity pool safety, concentration of holdings, project history, and overall risk profile.

(1) Market Traction

Bitget only lists assets that already have a robust user base and market traction. For projects debuting on the platform, Bitget generally examines:

• Number of potential airdrop participants (it must be over 4k);
• Number of potential IEO/IDO participants (it must be over 4k).

Example. Blur is an NFT marketplace that launched its token in February 2023. Shortly before launch, the platform had a trading volume of $33 million, the highest among NFT marketplaces. Bitget's on-chain analysis found that over 50,000 addresses met the Blur airdrop requirements, making it a key project for Bitget's debut listing. After sufficient preparation for listing, Bitget was the first to support and launch the token, meeting user demands and attracting many new users. For tokens that are already in circulation, Bitget evaluates on-chain metrics to consider their listing:

• Global 24-hour trading volume must exceed $1M;
• On-chain transaction count must exceed 30 per hour.

Example. Bitget listed the GAS token on October 25, 2023, as its trading volume surpassed $20 million and the price was showing a continuous upward trend. Bitget detected the movement and quickly supplemented it – after that, GAS's price rose from $2 to $29 in just two weeks.

(2) Community

Besides large trading volumes, gaining recognition from the community and market participants is essential. One of the ways to assess it is to look at what KOLs and top traders do – Bitget has established its own database of such traders, helping the exchange to discover the community and verify the on-chain metrics. Scraping Google Trends for the hottest daily searches is another tool that helps determine whether a project is truly being discussed and favoured. Additionally, some quantitative methods include (applicable to both debut and non-debut projects):

• Checking whether the project’s X (ex-Twitter) has over 10k followers, with ordinary posts having over 2,000 views;
• Ensuring the Discord/Telegram membership is over 5k, with 10+ effective community interactions per hour.

Example. ORDI is a token that passed this assessment and was listed on Bitget. The project conducted OTC trades on Discord before being listed on mainstream exchanges, and this activity and hype drove the community into a FOMO state. Bitget kept ORDI or radars, and after the project built its BRC-20 asset, it was quickly listed on Bitget.

(3) Technological Innovation

Genuinely innovative projects that solve market pain points likely contain growth potential. Bitget believes that the project's technological innovations should meet real market needs and be tightly integrated with other ecosystems to promote mutual advancement.

Example. UniBot is a Telegram bot allowing users to make fast token swaps on Uniswap DEX. The product meets real market needs by enabling automated trading in one of the world’s most popular messengers. After a careful assessment, Bitget concluded that the platform’s native token has growth potential – and became the first centralised exchange worldwide to list a Telegram Bot token, providing investors with solid returns.

(4) Token Economy

Strong tokenomics should accompany robust market traction, community engagement, and tech innovation to ensure the long-term sustainability of the project. Bitget has two separate frameworks for assessing the tokenomics of debut projects and already circulating coins.

For debut projects

• Matching the raised amount with the project's valuation. A key metric of any project’s economy is fully diluted valuation (FDV), calculated by multiplying token price by the total supply of the token. Generally, the FDV should not exceed the financing amount by more than 20X. For instance, a project that raised $5 million might be expected to have an FDV of around $50 million, but not exceeding $100 million.
• Assessing the reputation and track record of the investing institutions. If a project has a high fundraising amount but the investors are all below tier-2 or have unclear backgrounds, it will face skepticism and more rigorous due diligence.
• Examining token unlock schedule. If most of the team’s supply gets completely unlocked within 2 years or less, that might create selling pressure early on, and may also indicate that the project team lacks a plan and determination to continue operating.

For non-debut projects

For tokens already in circulation, Bitget analyses on-chain data to evaluate the project's economic health. One of the techniques involves comparing the token’s trading volume with its fully diluted valuation (FDV). For example, if a token has decent on-chain liquidity and an FDV over $10 million, but its 24-hour trading volume is below $1 million, Bitget will scrutinise the token more strictly because its FDV might be too high, potentially resulting in user losses.

According to the team’s observations, when the trading volume is low and the theoretical market cap does not align, the founding team may consider selling the token. This can cause a downward trend, eventually leading to trader losses.

A few days before the token generation event, spikes in trading volume often occur. If the token’s total value (FDV) exceeds the on-chain trading volume by more than 3X in this period, it can also be considered too high and require a more rigorous consideration.

(5) Security

The four methods above help identify promising projects – but then comes a security check that excludes a part of the potentially rewarding tokens. Strict control is essential to make sure no high-risk assets are listed on Bitget.

For initial listing projects

The stakes are especially high when it comes to nascent projects without a robust track record, so considerations here include all sorts of financial, security, compliance, political, ethical, and other risks. Bitget makes sure the project is not a Ponzi scheme, has nothing to do with gambling, the founding team doesn’t have a history of rug pulls, etc. Bitget evaluates contract security, liquidity pool safety, concentration of holdings, project history, and the overall risk profile.

Example. PulseChain was a blockchain project aiming to introduce a novel version of the Ethereum mainnet with a few improvements. Its PLS token was very hot on social media before launch, but the Bitget team audited the litigation between the PulseChain founder and the SEC and identified significant compliance risks. The token's listing was rejected.

For non-debut projects

Projects that have already listed their tokens in other exchanges are also subject to strict analysis. Bitget assesses risks related to smart contract security, token distribution, and many others. Issues indicating a high level of risk include suspended trading, the possibility for the contract issuer to change balances, potentially dangerous external calls in smart contracts, a buy/sell tax rate exceeding 20%, and many more.

When it comes to token distribution, projects where the team keeps 50%+ of the tokens are considered highly centralised and risky. Generally, addresses related to the token creator should not hold more than 20% of the supply. Here, “related addresses” do not only refer to the top obviously associated wallets: the Bitget team leverages blockchain explorers to track the token distribution and detect if the issuer disperses tokens into a large number of addresses. Some malicious token issuers use this method to achieve a seemingly dispersed token distribution, which opens doors to price manipulations, rug pulls, and rat trading.

Example. A few Simpson-themed tokens popped up as the memecoin market surged in 2024. While scrutinising one of them, Bitget discovered that it covertly created 800 addresses for itself, making the token distribution appear dispersed – while in reality, the vast majority of the supply was controlled by the issuer. Such tokens were identified and rejected by the Bitget team.

Once a token is listed, and more on-chain data become available, traders can follow these data as inputs to their investment process.

II. Blockchain Fundamental Metrics & Token Prices

By Nansen Research

A. On-chain scope

For its analysis of on-chain data vs governance token price changes, Nansen has scoped 12 chains. These are Ethereum and its layer-2 scaling chains: Arbitrum, Base, Celo, Linea, Polygon, Optimism, as well as other layers-1s, namely Avalanche, Binance Smart Chain (BSC), Fantom, Ronin, Solana, and Tron.

Two scopes were considered:

• One where Ethereum plus “scaling chains” are aggregated since these share the same economic and technological ecosystem
• One where all the 12 chains are kept separate.

Although the blockchains above share similar technological “building blocks”, they are used and valued for different purposes.

To go through some use cases: Ethereum and its scaling ecosystem represent the backbone of on-chain activity. Arbitrum and Optimism can be credibly benchmarked against each other as the two main optimistic roll-ups. Solana benefits from leading transaction speed, as well as positive reflexivity in the current bull cycle. Polygon covers several chains and technologies under its umbrella (zkEVM, validiums, plasmas, etc.). Tron is valued for payment use cases, notably in Asia, via USDT. BSC gains value from being associated with the Binance exchange.

B. What on-chain metrics?

Despite the variety in their use cases, a few fundamental characteristics are commonly sought after across chains:

• speed (often proxied by the number of transactions by a unit of time)
• cost of use (or revenue from the chain perspective) which is equivalent to transaction fees
• “network effect” or popularity, which can be roughly assessed by how much a blockchain is used (number of transactions) and by how many different users (number of wallets)

The metrics tested in this note were aggregated every week, and per chain: number of transactions, number of new wallets (having sent a transaction), fees in local gas currency, and in ETH. We added a popular metric among analysts, Total Value Locked (TVL) in ETH.

We considered and then discarded the number of blocks per week as it was not really comparable across chains. At the end of the analysis, we even included an off-chain metric, social sentiment towards the governance tokens of the chains in scope.

C. Visual Relationships

By charting governance token prices vs underlying chain metrics on two separate axes of the same graph, we can spot three categories of chain/ token patterns.

1) For the Ethereum plus L2 ecosystem (except for Celo), there appears to be a significant link between governance tokens and chains, whether the statistics are aggregated across this ecosystem and compared to ETH price, or by looking at chain-by-chain and token-by-token e.g. Optimism on-chain data vs OP price, Arbitrum on-chain data vs ARB price, Polygon vs MATIC, Ronin vs RON. Let’s note that this level of granularity was not realisable with Linea and Base as neither chain has issued a governance token. Outside of the EVM world, BSC metrics also seem to be moving with the BNB price, graphically.

2) Solana vs SOL presents some ambiguity for some metrics, notably for the number of transactions when charted vs SOL price. So does the couple Avalanche/ AVAX (strong relationship between price and on-chain metrics except for Fees).

3) The last group of chains does not have a visible consistent relationship between the on-chain metrics we selected and their governance token prices: Fantom vs FTM, Celo vs CELO, Tron vs TRX.

For a simplified view, we show only a few excerpts of the governance token price vs on-chain data graphs, to illustrate the three groups of patterns above:

Chart 1: New Unique Wallets on Ethereum + L2 Ecosystem vs ETH Price

Chart 2: Number of Transactions on Ethereum + L2 Ecosystem vs ETH Price

Chart 3: Fees (in ETH) on Ethereum + L2 Ecosystem vs ETH Price

Chart 4: TVL (in ETH) on Ethereum + L2 Ecosystem vs ETH Price

Chart 5: Number of Transactions on Solana vs SOL Price

Chart 6: TVL (in ETH) on Solana vs SOL Price

Chart 7: New Unique Wallets on Fantom vs FTM Price

D. Statistical Relationships

To supplement the visual observations above, we run a statistical test borrowed from Fama and MacBeth. The Fama-MacBeth regressions are widely used by financial practitioners to estimate the risk premia associated with equity market returns.

Why use this model in this instance? The on-chain data that we have selected for testing are “as close as it gets” to proxying chains’ economic fundamentals. We want to validate or invalidate the intuition that there is some statistical relationship between chain fundamentals and governance token prices. Likewise, the Fama-MacBeth factors have demonstrated some significance in explaining equity returns.

Now for the process: each statistic (number of transactions, number of new wallets, fees in local gas currency/ in ETH, TVL) represents a factor that we want to test as independent variables vs governance token prices as dependent variables. We start by running cross-sectional regressions per week where Token Price Weekly Return(at T) = (Num Tx % Chg Coeff 1 + Num New Wallets % Chg Coeff 2 + Fees % Chg Coeff 3 + TVL % Chg Coeff 4)(at T).

Cross-sectional means that we run the above equation across all the chains in our sample. The goal is to estimate all four coefficients every week and to obtain a weekly time series for each of these coefficients. We also try the equation above with all the combinations of subsets of factors.

The next step is to collect the time series for each coefficient (four time series), and average these across time by equal-weighting each week: this gives us four average coefficients or “risk features” across time. We run the t-test for each coefficient to test the significance of each as a regressor.

Finally, we go through the same process this time using price returns at (T+1) or (Week + 1) while keeping the independent variables at T. We want to verify whether our on-chain statistics can predict price returns. In that case, the average coefficients are called “risk premia”.

E. Fama-MacBeth Regression Results

In this section, we run through the results of step 2 of the regression methodology presented in section D. In a series of tables, we present average coefficients (risk features and risk premia) as well as their associated t-statistics. For a coefficient to be deemed significant with a confidence level of 95%, the calculated t-stat for a given coefficient must be superior or equal to 1.96 or inferior or equal to -1.96. In the tables below, the last column (on the right) features Newey and West t-stats: these are variations of t-stats adjusting for autocorrelation.

As you may recall from section A above, we have two sets of chain universes: one where we consider Ethereum and L2/ scaling chains as one ecosystem, and therefore aggregate on-chain variables for these, and one where we consider each L2 chain separately. The results of the tests that consider Ethereum and L2 as one ecosystem are much less significant than the results that proceed from testing chains separately. This makes intuitive sense: we found out earlier that Ethereum and L2 tended to have good visual fits between on-chain metrics and price graphs. Aggregating EVM chains gives more weight in the cross-sectional regressions to non-EVM chains.

Here, we will only present the results of the Fama-MacBeth regressions that consider each L2 chain and Ethereum separately. The starting date is each chain’s inception week.

We start by regressing price returns on all four metrics in scope, in a contemporary way (all % change at time or week T). As Table 1 illustrates, only Fees in ETH and TVL in ETH have absolute t-stats above the threshold of 1.96.

Table 1: Results Contemporary Fama-MacBeth Regressions: Price Return vs (Fees in ETH, New Active Wallets, Num Transactions, TVL in ETH) % change

To improve the model, we exclude the Number of New Active Wallets and the Number of transactions, the metrics with non-significant results, and re-run the two-stage regressions for contemporary time series. This time, the model is satisfactory as all coefficients meet our critical threshold of 1.96 (see Table 2 below).

Table 2: Results Contemporary Fama-MacBeth Regressions: Price Return vs (Fees in ETH, TVL in ETH) % change

Staying with these two variables, we want to find out whether TVL in ETH and Fees in ETH weekly % changes are predictive of price returns, one week in advance. The results are presented in Table 3. As can be observed, t-stats remain significant for TVL in ETH but not so for Fees in ETH.

Table 3: Results Leading Fama-MacBeth Regressions: Price Return vs (Fees in ETH, TVL in ETH) % change

Finally, we stick with Fees in ETH and TVL in ETH but split them apart to run four one-factor models, one contemporary and one leading, per factor.

The findings, presented in Tables 4 to 7, are interesting: moving from two-factor to one-factor confirms the strong contemporary relationships of fees with prices, and TVL with prices, respectively (tables 4 and 6). Interestingly, Fees and TVL do a better job of predicting price returns if they are split into two one-factor models (see Tables 5 and 7).

Table 4: Results Contemporary Fama-MacBeth Regressions: Price Return vs (Fees in ETH) % change

Table 5: Results Leading Fama-MacBeth Regressions: Price Return vs (Fees in ETH) % change

Table 6: Results Contemporary Fama-MacBeth Regressions: Price Return vs (TVL in ETH) % change

Table 7: Results Leading Fama-MacBeth Regressions: Price Return vs (TVL in ETH) % change

To summarise our regressions’ results, TVL in ETH and Fees in ETH, together, form the best model for changes in governance prices at the same time T. TVL in ETH and Fees in ETH are significant risk features of price.

As for predicting price returns, one week in advance, “TVL in ETH” is a significant risk premium in a one-factor model and so is the metric “Fees in ETH”. Both have positive risk premia or coefficients, meaning that higher fees and higher TVL tend to be associated with higher subsequent returns.

As a side note, replacing Fees in ETH with Fees in local gas currency leads to similar results.

F. Fama-MacBeth Regressions including off-chain data

Encouraged by our on-chain findings, we decided to add off-chain variables to our factors. These variables have as input the messages and reactions we collect from our Nansen Alpha Discord, where analysts and Alpha clients interact. Messages are parsed for mentions of Token Names, and the associated sentiment is analysed using Artificial Intelligence. For a complete overview of the methodology of the Alpha Discord Token Parser, please see this article (only accessible to Nansen Alpha clients).

From parsed token mentions and sentiment, we extract six metrics, aggregated per governance token of the chains in the scope of this analysis and per week. The first three metrics are the number of messages, the number of users (or separate Alpha Discord accounts), and the number of reactions (e.g. emojis). The other three metrics calculate sentiment (positive 1, neutral 0, negative -1) associated with each token and weigh these by the number of messages, number of users, and number of reactions, respectively.

Chart 8: Nansen Alpha Discord Message-Weighted Sentiment, per chain, per week

Chart 9: Nansen Alpha Discord User-Weighted Sentiment, per chain, per week

As you can observe from charts 8 and 9 above, sentiment data are only available from the first week of January 2023 to the end of March 2024, which represents a smaller time sample than in our prior analysis.

We create indices of mentions and sentiments where each of the six metrics equals 100 on the first week of January 2023 and then we calculate a weekly rate of change. We regress token price weekly returns against these rates of changes.

Before considering our off-chain regressors, we return to TVL in ETH and Fees in ETH in this smaller time sample of one year and three months. The results are noticeably less significant than in our larger time-sample analysis, even for contemporaneous relationships (table 8).

Table 8: Results Contemporary Fama-MacBeth Regressions: Price Return vs (Fees in ETH, TVL in ETH) % change, January 2023 to March 2024

Moving to our six Mention and Sentiment Alpha Discord metrics, we find that two of the six metrics: Sentiment weighted by number of messages and Sentiment weighted by number of users, deliver decent results.

Table 9: Results Contemporary Fama-MacBeth Regressions: Price Return vs (Fees in ETH, Message-weighted Alpha Discord Sentiment) % change, January 2023 to March 2024

Table 10: Results Leading Fama-MacBeth Regressions: Price Return vs (Fees in ETH, Message-weighted Alpha Discord Sentiment) % change, January 2023 to March 2024

Table 11: Results Contemporary Fama-MacBeth Regressions: Price Return vs (Fees in ETH, User-weighted Alpha Discord Sentiment) % change, January 2023 to March 2024

Table 12: Results Leading Fama-MacBeth Regressions: Price Return vs (Fees in ETH, User-weighted Alpha Discord Sentiment) % change, January 2023 to March 2024

The t-stats are better for these two Sentiment metrics than for the Fees risk premium since January 2023 (tables 9 and 11), but, at about 1.21 and 1.26 their Newey and West t-stats score below our 95%-significance threshold for a contemporary relationship with price changes.

Looking at leading relationships, we cannot find sufficient t-stats (tables 12 and 13). There could be several reasons behind this lack of satisfactory results: we suspect that a higher time granularity than weekly is required when using social sentiment data, as price moves occur very fast.

The difficulty of finding significant predictors of crypto tokens price returns highlights the necessity for crypto market participants to combine many inputs for decisions, spanning quantitative on-chain and off-chain data, and qualitative criteria.