On- and Off-Chain Demand and Supply Drivers of Bitcoin Price

1 Introduction

Cryptocurrencies are exchanged between owners in many different ways. Depending on the platform used, crypto transactions can be grouped into two broad types: “on-chain” and “off-chain”, where “chain” refers to a sequence of blocks. Most of the existing cryptocurrency literature has investigated “on-chain” transactions – mainly due to data availability considerations – even though most of the cryptocurrencies are being exchanged “off-chain” (Tierno 2023; ESMA 2024). The impact of off-chain transactions on cryptocurrency prices has been considerably less studied (Cerutti et al. 2024). This paper attempts to address this gap by examining the relationship between both on- and off-chain demand and supply-side factors and cryptocurrency prices.

“On-chain” crypto-asset transactions take place directly on the blockchain. They involve a transfer of crypto-coins from one digital wallet address to another, and they are recorded on blockchain, a distributed ledger. In distributed ledger technologies, data are structured into blocks and each block contains records of a transaction or a bundle of transactions. The transaction validation is based on a consensus mechanism (Proof of Work in the case of Bitcoin), decentralisation (without intermediaries or a central authority) and cryptography (public-private key encryption). This ensures trustworthiness (users do not need to trust a central authority or intermediary), security (transactions cannot be altered once recorded), and transparency (all transactions are publicly visible and traceable) (Mount 2020; Soares 2022; Tierno 2023; ESMA 2024). On-chain transactions may include transactions related to the purchase of goods and services (e.g. wallet-to-wallet transactions, store of value), decentralised finance (DeFi) transactions (e.g. decentralised exchanges (DEXs), lending/borrowing) and other blockchain-based economic activities (e.g. gaming, supply chain management and traceability, non-fungible tokens).

The rapid growth of on-chain transaction data – often referred to as blockchain bloat – presents an increasingly pressing challenge for the blockchain technology (Alzoubi and Mishra 2024). One prominent response has been the growing reliance on crypto-asset transactions that take place “off-chain”. Off-chain transactions refer to all other crypto ownership changes that are not recorded on the blockchain. Typically, in an off-chain transaction, the legal ownership of a crypto asset changes but it remains associated with the same digital wallet (e.g. the wallet of a cryptocurrency centralised exchange (CEX)) and the ownership change is not recorded on the blockchain. Instead, off-chain transactions are recorded on centralised ledgers or private order books of intermediaries, such as crypto exchanges, custodial wallets, and financial institutions. Survey data from Blandin et al. (2020) shows that, in terms of both volumes and numbers, off-chain transactions continue to be dominated by fiat-crypto asset trades (and vice versa), meaning that users primarily interact with ‘gateway’ service providers, such as exchanges, to enter and leave the crypto-asset ecosystem. The most prominent examples of off-chain transactions are those carried out on CEXs such as Binance or Coinbase.^[1]

The main advantages of off-chain transactions are lower costs, faster transaction execution and the ability to address privacy concerns (anonymity) (Mount 2020; Soares 2022; Tierno 2023; Alzoubi and Mishra 2024; ESMA 2024), as off-chain transactions reduce the activity burden on the blockchain. Off-chain transactions include, but are not limited to, transactions on CEXs related to the buying and selling of crypto coins, purchasing goods and services (e.g. on the Lightning Network), and financial transactions (e.g. lending/borrowing, margin trading). They also include transfers of crypto-asset ownership on decentralised layer 2 protocols. Layer 2 protocols are secondary decentralised networks built on top of the primary blockchain to achieve greater scalability (handling a high volume of trades), faster transaction times, and lower transaction costs. After processing transactions on layer 2, they are aggregated for a final settlement on the main blockchain (layer 1) (Soares 2022; Tierno 2023; Alzoubi and Mishra 2024; ESMA 2024).

Most existing empirical analyses of crypto-assets rely on data generated by on-chain activity (“on-chain data”) (e.g. Ciaian et al. 2016; Li and Wang 2017; Gbadebo et al. 2021; Cerutti et al. 2024). Analyses using on-chain data are useful, among others, for understanding the level of activity in the blockchain-based Bitcoin economy. For example, the volume of on-chain transactions reflects the level of adoption of the blockchain-based crypto economy. An increase in transaction volume is associated with higher network traffic velocity, a larger user base, higher trading activity, and more generally, a greater trust in the blockchain-based crypto economy. Conversely, a decrease in transaction volume may signal uncertainty among users and indicate a lower level of the overall adoption of the blockchain-based economy (OSL 2025). On-chain data can also provide insights into the proportion of illicit activity involving cryptocurrencies, which is estimated at 0.34 % of the total transaction volume (Chainalysis 2024). Furthermore, on-chain analysis can be used to deduce the value of crypto-assets being moved between real-world entities on the blockchain. For instance, it has been shown that exchanges account for 90 % of all funds sent by crypto-asset services (Blandin et al. 2020).

Despite their merits, layer 1 blocks on distributed ledgers do not contain information about off-chain sales, which are recorded in private order books of intermediaries such as cryptocurrency exchanges or financial institutions. As on-chain data exclude purchases of crypto-assets with fiat currency, sales of crypto-assets for fiat currency, and swaps between crypto-assets, they provide an incomplete and potentially biased picture of crypto markets. To gain a more comprehensive understanding of transactions on different blockchain layers, analyses leveraging off-chain data are necessary. Indeed, a growing number of studies are incorporating off-chain data into their empirical analyses of crypto-assets (e.g. Bouri et al. 2017; Giudici and Abu-Hashish 2019; Szetela et al. 2020; Koutmos 2023; Guo et al. 2025). However, while a large and vibrant body of literature has focused primarily on on-chain trading and, to a lesser extent, on off-chain trading of crypto-assets, empirical evidence combining on- and off-chain investor behaviour, their aggregate market-level consequences, and their impact on crypto-asset returns remains limited. This paper seeks to address this gap by examining both on- and off-chain demand and supply drivers in order to identify their relationship with the Bitcoin price dynamics. Building on conceptual and transaction pattern analyses, the paper formulates testable hypothesis and uses time-series analysis of daily data from 2019 to 2024 to estimate how on-chain and off-chain supply and demand drivers are correlated with short- and long-run movements in the Bitcoin price.

The rest of the paper is structured as follows. Section 2 develops the research hypothesis based on insights from the existing conceptual literature and empirical observations. Section 3 describes the econometric methodology to be used to analyse the short- and long-term relationships between Bitcoin price dynamics and on- and off-chain drivers. Section 4 presents the data sources, the construction of the variables and the descriptive statistics. Section 5 discusses the empirical results. Finally, Section 6 summarises the main findings, outlines policy and market implications, and highlights the main limitations of the paper.

2 Research Hypothesis

We approach the hypothesis from both conceptual analyses and transaction patterns perspectives to provide a comprehensive analytical framework.

2.2 Transaction Pattern Analysis

To develop a hypothesis that can be tested empirically using time-series mechanisms, we further examine Bitcoin transaction patterns and relate on-chain versus off-chain demand and supply factors to the Bitcoin price. Demand factors are those that increase the relative attractiveness to buy or hold Bitcoin (e.g. on-chain Bitcoin transactions, fiat liquidity on CEXs), while supply factors are those that augment the number of available coins (e.g. total issued Bitcoins). On-chain demand and supply variables capture transactions that occur directly on the blockchain. These include transactions involving the purchase of goods and services, decentralized finance (DeFi) transactions, and other blockchain-based economic activities. Off-chain demand and supply variables, on the other hand, capture transactions that occur outside the blockchain, such as transactions involving the buying and selling of Bitcoin with fiat currency on CEXs or over-the-counter (OTC) markets.

At least three factors may explain why the fundamentals of cryptocurrency trading differ between on-chain and off-chain transactions, leading on-chain crypto-coin users and off-chain crypto-asset traders to respond differently to the same market signals and ultimately affecting crypto prices differently. First, the available evidence suggests that the primary purpose of off-chain transactions serve a fundamentally different purpose than on-chain crypto-asset transactions (Cerutti et al. 2024; Feyen et al. 2022; Makarov and Schoar 2021; Thakkar et al. 2024; Tierno 2023). Second, key statistical properties, such as mean and variance, exhibit significant structural differences between on-chain and off-chain transactions. Third, off-chain transactions account for the majority of total Bitcoin transaction volume (Figure 1), suggesting that they may have an impact on crypto-asset prices.

Figure 1:

Global monthly Bitcoin on-chain and off-chain transaction volumes and on-chain share, 2020–2024. Source: authors’ computations based on on-chain transaction volume data from theblock.co and off-chain trading volume data from ccdata.io. Notes: Bitcoin transaction volume (bar chart) is measured in trillions USD at current prices on the left y-axis; Bitcoin on-chain share (line chart) is measured as a percentage of Bitcoin transaction volume on the right y-axis.

First, as discussed above, on-chain transactions involve the direct transfer of cryptocurrencies between wallets and the execution of various decentralised economic operations directly within the blockchain network itself. As such, these transactions serve as a key indicator of the market fundamentals of the decentralised cryptocurrency economy (Böhme et al. 2015). These transactions can be conducted for various economic purposes, such purchasing goods and services, making cross-border transfers, conducting long-term holding, carrying out decentralised financial transactions, facilitating illicit trade transactions, distributing block rewards and others (Makarov and Schoar 2021; OSL 2025). However, not all on-chain Bitcoin transactions are necessarily linked to economically relevant activities. Makarov and Schoar (2021) found that, from 2017 to 2020, around 90 % of on-chain Bitcoin transaction volume consisted of spurious transactions in which entities transferred Bitcoin between themselves, similar to moving money between pockets. The remaining 10 % of on-chain transactions represented genuine transactions between different parties.

Bitcoin trading is one of the key activities that directly affects the Bitcoin price dynamics as it is an outcome of the interaction between supply and demand factors. Most Bitcoin trading occurs off-chain on CEXs rather than DEXs (Figure 1). According to CoinMarketCap.com, a website that tracks the performance of various cryptocurrencies, DEX trading volume represents less than 1 % of the total daily Bitcoin trading volume (CoinMarketCap 2024).^[2] Off-chain trading on CEXs is generally considered speculative, involving transactions aimed at extracting gains from price movements or hedging against alternative investments (e.g. stocks, commodities) rather than sustaining economic activities like purchasing goods and services (Kukacka and Kristoufek 2023; ESMA 2024; Ozer et al. 2024). Hougan et al. (2019) report that 95 % of Bitcoin transactions on CEXs are linked to the buying and selling of Bitcoin with no economic value, most of which is driven by fake or wash trading. This suggests that off-chain transactions are predominantly speculative. This is expected to be less the case for on-chain Bitcoin transactions, which are more likely to be influenced by market fundamentals linked to the decentralised cryptocurrency economy (Ciaian et al. 2016) because on-chain trading on DEXs is much less frequent.

Second, Cerutti et al. (2024) found that on-chain transactions, on average, differ significantly from off-chain transactions. For instance, on-chain transactions tend to be substantially larger than off-chain transactions (LocalBitcoins). The average on-chain transaction size is 13.35 Bitcoin, while the average off-chain transaction size is only 0.018 Bitcoin. Similarly, at the same Bitcoin price, the maximum on-chain transaction reaches US$300 million, compared to US$1.875 million for off-chain transactions. These differences in transaction size and distribution suggest that different types of market participants operate in on-chain and off-chain environments (Cerutti et al. 2024). The IMF provides evidence that the primary motivations for cross-border off-chain transactions are circumventing capital flow restrictions and facilitating remittance transfers. The descriptive statistics in Table 3 further support this, showing significant structural differences in data moments, such as the mean and variance, between on-chain and off-chain drivers. For example, on-chain Bitcoin demand-side drivers – as measured by On-chain BTC transactions variables – are fundamentally different from off-chain Bitcoin demand-side drivers – as measured by Bank netflow, Bank reserve, and Fund volume variables. Importantly, these differences between on-chain and off-chain drivers seem to be structural, reflecting distinct trading motivations and purposes (Cerutti et al. 2024).

Third, the impact of off-chain drivers on crypto-asset prices can be significant, depending on the proportion of off-chain transactions relative to on-chain transactions in the total crypto-asset trading volume. Figure 1 illustrates the scale of off-chain activity relative to on-chain transactions by showing global monthly Bitcoin transaction volumes: on-chain transactions (dark bars) and off-chain transactions (light bars). These are measured in trillions of USD at current prices (left Y-axis) and are based on data from theblock.co and ccdata.io. Figure 1 reveals several key insights: (i) The ratio of on-chain to off-chain transactions has fluctuated considerably over time with the on-chain share dropping below 15 % at its lowest and exceeding 55 % at its highest. (ii) Over the last two years analysed, the ratio of on-chain to off-chain transactions has remained relatively stable, ranging from 15 % to 25 %. (iii) Overall, off-chain transaction volumes appear substantially larger than on-chain volumes. Our estimates suggest that the ratio of off-chain to on-chain volume has been around 5:1 over the last two years.

This estimated ratio aligns closely with that of Feyen et al. (2022), who found a similar off-to-on-chain volume ratio of approximately 6:1. Similarly, Makarov and Schoar (2021) estimate that, since 2015, approximately 75 % of the total real Bitcoin volume has occurred off-chain via exchanges or exchange-like entities (e.g. online wallets, large institutional traders and OTC desks) implying an off-to-on-chain ratio of 4:1. Between 2017 and 2020, weekly on-chain genuine transaction volumes typically ranged from 50,000 to 160,000 Bitcoins, while weekly off-chain volumes ranged from 100,000 to 300,000 Bitcoins (Makarov and Schoar 2021).

In summary, based on the conceptual and empirical analyses presented above, it is hypothesised that off-chain demand and supply-side drivers will exert a distinct influence on Bitcoin prices and will be related to them differently than on-chain drivers.

Hypothesis: Off-chain demand and supply factors are related to the Bitcoin price dynamics due to their dominant share in the total transaction volume and their speculative nature.

Although on-chain transactions constitute a minority of total Bitcoin trades, they may nevertheless be relevant – and potentially significant – in determining the price of Bitcoin. As evidenced above, the qualitative and structural differences between on-chain and off-chain transactions suggest that on-chain data can capture unique insights into investor behaviour and user engagement in the decentralised Bitcoin economy (e.g. speculation versus fundamentals). These insights may, in turn, have ramifications for the price dynamics. Additionally, a substantial body of research has been dedicated to estimating the relationship between on-chain drivers and Bitcoin price formation. While the findings in the literature are mixed, several studies have concluded that some on-chain drivers, alongside macro-financial variables, may also play a role in determining the Bitcoin’s price (e.g. Ciaian et al. 2016; Li and Wang 2017; Nguyen et al. 2019; Gbadebo et al. 2021; Madichie et al. 2023). Hence, including on-chain variables in the estimable equations provides a link to the existing literature. Further, this approach allows us to empirically evaluate the relative importance of off-chain and on-chain supply and demand factors in the Bitcoin price formation. Finally, the available evidence suggests that focusing solely on either off-chain or on-chain data can lead to an incomplete understanding of the Bitcoin price formation (Cerutti et al. 2024). Therefore, we focus on both types of transactions in our estimations.

3 Methodology

We employ an autoregressive distributed lag (ARDL) model to identify the differentiated relationship between on-chain and off-chain supply and demand drivers and the Bitcoin price. The choice of the ARDL model over alternative estimation methods is based on several considerations. First, it is particularly advantageous in capturing both short- and long-run relationships among variables regardless of whether they are stationary at levels I(0), first differences I(1), or cointegrated. Second, it accommodates different lag lengths for various regressors. Third, assuming that the price is exogenously determined by supply and/or demand variables, the estimates obtained are considered unbiased and efficient, as the approach mitigates issues related to serial correlation and the endogeneity problem in a single-equation framework (Madichie et al. 2020, 2023; Pesaran et al. 2001). Moreover, in the context of cryptocurrencies, their supply is fully exogenous as information about Bitcoin growth is publicly available to all market participants ex-ante.

Given these advantages, the ARDL methodology has been widely employed to study the cryptocurrency price formation (e.g. Gamal et al. 2019; Bouraoui 2020; Stoian and Iorgulescu 2020; Nguyen et al. 2025; Atmaca and Karadaş 2020; Wang et al. 2023). This approach has been adopted for general analyses of crypto prices (e.g. Hernandez et al. 2021; Nouir and Hamida 2023; Aliyev and Eylasov 2025; Gaies et al. 2021; Sifat et al. 2019) and for analyses that explicitly consider supply and demand factors (e.g. Ciaian et al. 2016; Nguyen et al. 2019; Gbadebo et al. 2021; Madichie et al. 2023; M’bakob 2024).

We acknowledge that endogeneity may arise if cryptocurrency prices are not determined exogenously by supply and/or demand variables. This is because the considered demand and supply drivers of the Bitcoin price may be simultaneously determined, leading to a simultaneity bias if not properly accounted for (Thurman 1986). The ARDL approach helps addressing simultaneity bias from a dynamic feedback by incorporating lagged regressors, which mitigates the risk of simultaneity bias over time. By including lagged dependent and independent variables, the ARDL approach also helps to mitigate endogeneity resulting from a reverse causality and past feedback effects. However, we note that the ARDL approach may not fully address contemporaneous simultaneity bias from all sources of endogeneity, which is a promising avenue for future research.

As usual, we start with investigating the presence of a long-run relationship between time series using the ARDL bounds test introduced by Pesaran et al. (2001). The ARDL method can be applied to stationary time series at I(0) or I(1) and to cointegrated time series. However, Ouattara (2004) points out that the F-statistics provided by Pesaran et al. (2001) become invalid if the model includes I(2) variables. To ensure that none of the time series are integrated of order I(2) or higher, we use the Augmented Dickey-Fuller (ADF) and the Phillips-Perron (PP) tests to assess the stationarity of the data series and their first differences. We use the Akaike Information Criterion (AIC) to determine the optimal number of lags. The general form of the ARDL (p, q) model reads is as follows:

where, c ₀ and c ₁ t are the intercept and a linear trend, respectively; y is the dependent variable (Bitcoin price); x is a vector of independent variables (demand and supply drivers related to different types of Bitcoin transactions, macro-financial variables); p refers to the number of optimal lags of the dependent variable; q is the number of optimal lags for each explanatory variable; and u _t is a white noise error term. We include a set of exogenous variables, z _t , with predictive power to better explain the short-run deviations in y without affecting its equilibrium (Kripfganz and Schneider 2023).

The ARDL bounds testing technique is used to check for a long-run relationships. This involves calculating F- and t-statistics and comparing them to critical value bounds. Pesaran et al. (2001) have suggested two types of critical values for a given significance level. One type assumes that all the variables are I(1), and the other type assumes that all the series are I(0). If the calculated F- and t-statistics are below the lower bounds, the null hypothesis of no long-run relationship cannot be rejected. This indicates that an ARDL model in first differences, without an error correction term, should be estimated. If the F- and t-statistics are found between the lower and upper bounds, the outcome is inconclusive. Conversely, if the F- and t-statistics cross the upper bound, the null hypothesis of no cointegration can be rejected. In this case, the error correction model to be estimated is (Hassler and Wolters 2006):

where θ denotes the long-run coefficients, ψ denotes short-run multipliers, and α denotes the speed at which the dependent variable adjusts to a short-run shock, which indicates the rate at which the variables revert to their long-run equilibrium.

As suggested by Pesaran et al. (2001), we performed a series of diagnostic tests to validate the ARDL results to confirm that the models satisfy the standard assumptions of normally distributed error terms, the absence of serial correlation and heteroscedasticity, and stable coefficients. Specifically, we used the Breusch-Godfrey LM test and Durbin’s alternative test for autocorrelation, the Breusch-Pagan/Cook-Weisberg test for heteroscedasticity, normality testing, and the cumulative sum test for parameter stability.

To determine the appropriate lag structure for our ARDL models and ensure parsimonious, well-specified equations, we employ an automated model selection process in Stata. Consistent with the literature, the optimal number of lags for both the dependent and independent variables in each model specification was chosen based on the AIC. The model with the lowest AIC value is considered the most efficient, as it strikes the best balance between a good fit and model simplicity, thereby avoiding over-parameterization. The final model specifications presented in our results were selected from a pool of up to 10,000,000 potential combinations of lags, guided by the model with the minimum AIC value. This systematic approach ensures that our final models are not arbitrarily complex but are instead empirically grounded and parsimonious.

Although the ARDL approach is well-suited for modelling cointegration and short- and long-run temporal dynamics between variables with different integration orders, it is important to keep in mind its limitations. While the adopted ARDL framework identifies significant associational relationships between the Bitcoin price and its on-chain and off-chain drivers, it does not establish a definitive causality. As mentioned earlier, the bidirectional relationship between price and transaction activity – wherein trading volume influences price changes and vice versa – introduces the possibility of endogeneity and simultaneity bias. This is a common challenge in financial time series analysis, particularly when using high-frequency data from centralized and decentralized markets. Accordingly, our estimation results should be interpreted as revealing strong statistical associations and dynamic influences rather than strict causal impacts.

To mitigate concerns about potential endogeneity and simultaneity bias and to ensure the robustness of our results, we extend our empirical strategy beyond the baseline ARDL specification. First, we test for the weak exogeneity of explanatory variables by including the error-correction term (ECM) in auxiliary regressions. Next, we estimate instrumental variable (IV) models that are operationalised using the ivreg2 command in Stata. We employed lagged values of the potentially endogenous regressors as instruments. We then conducted diagnostic tests (underidentification, weak identification, and Hansen J) to assess the validity and strength of the instruments. Furthermore, to mitigate possible finite-sample bias arising from weak instruments, we complement the standard two-stage least squares (2SLS) estimates with the Fuller(1) estimator.

4 Data and Variable Construction

For the empirical analysis, we use daily data for the period 04/12/2019-25/01/2024. A detailed summary of the data used in the estimations and their sources is provided in Table 1. Table 2 provides descriptive statistics of the time series. In all specifications, the dependent variable is the price of Bitcoin, expressed in US dollars per Bitcoin. In line with the tested hypothesis, we include several explanatory variables to proxy the demand and supply drivers of on-chain and off-chain Bitcoin transactions. The selection and construction of off-chain variables is strongly influenced by the limited data availability. Indeed, this is one of the few analysis to date that make use of data on transactions that are not recorded on layer 1, and are therefore not available in the publicly shared blockchain information. Furthermore, while the use of high-frequency daily data and the integration of off-chain variables are essential, we acknowledge that several off-chain proxies, such as Bank reserve and Fund volume, have not yet been standardised in the academic literature and hence require an additional explanation.

Regarding the off-chain demand-side, we consider three alternative explanatory variables: Bank netflow, Bank reserve and Fund volume. Since these variables represent the demand-side, it is expected that they will have a positive relationship with the price of Bitcoin. Bank netflow measures the net amount of Bitcoin flowing into and out of the digital asset banks that provide various financial services including lending, custody, staking, payments, and synthetic assets (e.g. stablecoins or tokenised assets). This metric provides insights into the overall Bitcoin demand. A positive net flow suggests an increasing demand, as more investors deposit Bitcoin into trading platforms, potentially for long-term holding. Conversely, a negative net flow suggests reduced demand, indicating that more investors are withdrawing Bitcoin from these platforms, possibly to hold in private wallets or for other purposes. Bank reserve represents the USD value of coins held by the digital asset banks and indicates their level of liquidity. A larger Bank reserve indicates greater liquidity and, therefore, demand potential, as more funds are readily available to facilitate transactions. Theoretically, this variable is linked to demand through the concept of institutional liquidity: a rising reserve indicates a growing pool of ready capital within digital asset banks, signalling demand pressure from investors preparing to deploy funds. Fund volume refers to the driver of off-chain transactions associated with regulated Bitcoin funds, such as trusts, exchange-traded funds (ETFs) and mutual funds. It measures the trading volume of Bitcoin in regulated funds. An increase in fund volume signals higher liquidity and stronger activity of regulated funds, which may put upward pressure on the Bitcoin price. This variable is connected to demand through its function as a signalling mechanism: an increase in the trading volume of regulated funds is expected to demonstrate a collective increase in investor confidence and capital allocation from institutional sources, stimulating market demand in the process.

Regarding the off-chain supply-side, we consider two alternative explanatory variables: Exchange netflow and Exchange reserve. These variables are expected to be negatively correlate with the Bitcoin price. Exchange netflow refers to Bitcoin supply drivers as it reflects the movement of Bitcoin into and out of exchanges, directly influencing the supply dynamics in the CEX market. An increase in Exchange netflow suggests that more Bitcoin is being moved to exchanges. This can indicate a potential selling activity and exert downward pressure on the Bitcoin price. Exchange reserve refers to the total amount of Bitcoin held in CEXs. It indicates the total amount of Bitcoin supply that is readily available for trading. A higher reserve means more Bitcoin is available on exchanges, which could potentially increase the volume of coins offered.

On-chain data are standard in the literature and hence their descriptions are more concise. The variable On-chain BTC transactions proxies for on-chain demand-side transactions of Bitcoin. It is calculated by subtracting the transactions flowing into or out of exchanges from the total on-chain Bitcoin transactions.^[3] A higher value of this variable indicates a higher intensity of non-exchange-related Bitcoin activities (non-DEX), which are expected to be positively related to the Bitcoin price.

The on-chain supply-side variables considered in the estimations include Total supply and Coin days destroyed. These variables are expected to have a negative relationship with the Bitcoin price. Total supply measures the total number of issued (minted) Bitcoins. It measures the cumulative amount of all Bitcoins created since Bitcoin’s inception, thus provides an indicator of the circulating supply of Bitcoins. Coin days destroyed is calculated by multiplying the number of Bitcoins in transaction by the number of days since those coins were last spent. An increase in this variable suggests that long-term Bitcoin holders (i.e. holders of inactive coins) are liquidating their positions, potentially exposing their holdings to selling pressure.

In reality, the distinction between the on-chain and off-chain drivers may not be as straightforward as the above binary classification suggests, however. This is particularly evident with regulated fund flows, which often span both domains. For instance, the increasing popularity of Bitcoin ETFs has resulted in a substantial concentration of the Bitcoin supply under the control of regulated institutions such as BlackRock and Fidelity. Although these ETFs are traded off-chain, the underlying asset is held on-chain. This creates a dual-layered market, with the on-chain supply increasingly held in large, consolidated addresses while day-to-day trading and price discovery for these assets occurs entirely off-chain. Additionally, the variables used in our analysis do not directly capture investor motives, such as speculative behaviour or investment driven by knowledge of the cryptocurrency ecosystem. Instead, our conjecture of off-chain transactions as speculative and on-chain transactions as aligned with fundamental utility is indirect and based on the theoretical and empirical arguments presented in Section 2.

Following the literature (Apergis 2025; Ozer et al. 2024), we also included macro-financial developments as control variables in the estimated equations. Bitcoin is often considered as an investment asset, with potential investors weighing the expected benefits of investing in Bitcoin against other assets or using it as an inflation hedge (Ciaian et al. 2016; Sören 2023; Cong et al. 2024). As a result, macro-financial developments (e.g. the stock market, inflation, the gold price) are also expected to be related to the price of Bitcoin (Apergis 2025; Ozer et al. 2024). We consider the following macro-financial variables: the Federal Funds Effective Rate (DFF), the Market Yield on U.S. Treasury Securities (DFII10), the Consumer Price Index (CPIAUCSL), the Wilshire 5000 Price Index (WILL5000PR) and the Gold price. The DFF is the US Federal Reserve’s main tool for influencing monetary policy. Changes in the DFF reflect the stance of the US Federal Reserve on monetary policy. Decreasing rates indicate an expansionary approach to stimulate economic activity and inflation, while increasing rates signal a more restrictive approach aimed at controlling inflation. DFII10 represents the real yield, or real interest rate, on U.S. Treasury securities with a 10-year maturity, adjusted for changes in inflation as measured by the Consumer Price Index (CPI). CPIAUCSL is a measure of inflation of USD. The WILL5000PR represents the market value of all American stocks actively traded in the United States. The Gold price represents the price of a commodity asset – typically considered as a store of value by investors.

Since Bitcoin is often considered a store of value asset (Yae and Tian 2024), a negative relationship between the Federal Funds Effective Rate (DFF) and Bitcoin prices is expected: the Bitcoin price is expected to increase when monetary policy is expansionary (low DFF) and decrease when monetary policy is contractionary (high DFF). Similar holds for CPIAUCSL. In the presence of higher inflation (high CPIAUCSL), the Bitcoin price is expected to increase if Bitcoin is perceived as a store of value asset. Potential investors also often consider Bitcoin as an alternative investment opportunity among many other possible investment opportunities (such as stocks, treasury bonds). Since Bitcoin competes with other financial assets for investors’ attention, it must deliver competitive expected returns. The return arbitrage between alternative investment opportunities implies a positive price relationship between the price of Bitcoin and financial assets (i.e. DFII10, WILL5000PR) (Apergis 2025; Ciaian et al. 2018; Ozer et al. 2024). DFII10 serves as a proxy for the risk-free investment alternative, while WILL5000PR represents riskier investment alternatives. The Gold price is expected to be positively related to the Bitcoin price because gold may represent a less risky investment opportunity or an alternative store of value.

All variables are treated as endogenous in the estimations, with the exception of Total supply (Table 1). The variable Total supply is considered to be exogenous because it is predetermined by a fixed decreasing supply Bitcoin algorithm, is publicly known and is expected to affect the dependent variable without being affected by it.^[4]

To account for structural changes in the Bitcoin market development over time, we have included different dummy variables in the estimated models (Table 1). Dummy1 takes the value 0 before 20/11/2020 and 1 after this date. It takes into account the implications of the monetary and fiscal stimulus measures related to the Covid-19 pandemic adopted by different countries. Dummy2 takes the value 0 before 8/11/2022 and 1 after this date, corresponding to the collapse of the FTX cryptocurrency exchange. First, we run estimations with these two dummy variables. For robustness, we also re-estimate all models by considering Dummy3, which takes the value 1 between 20/11/2020 and 8/11/2022 and 0 outside this period.

5 Results

6 Conclusions

This paper investigates the relationship between different segments of the Bitcoin market (on-chain versus off-chain) and the Bitcoin price in the short- and long-run. After detailing the conceptual and transaction patterns perspectives of the blockchain and off-chain activity, we develop testable hypothesis associating on-chain and off-chain demand and supply factors to the Bitcoin price. The hypothesis posits that both off-chain and on-chain demand and supply drivers are related to the price of Bitcoin. As usual, we include controls for macro-financial developments to account for broader economic factors that may be related to the Bitcoin price dynamics. To empirically investigate this question, we apply time-series analytical mechanisms to daily data from 2019 to 2024.

Our estimates suggest that, in the short-run, both off-chain demand and supply drivers are significantly related to the Bitcoin price. This is fully in line with the tested hypothesis. In the long-run, however, we found only a partial support for the tested hypothesis: off-chain demand drivers exhibit a statistically significant relationship with the Bitcoin price, whereas off-chain supply drivers do not. As conceptual and transactional analyses of Bitcoin suggest, off-chain transactions tend to reflect the activity of less informed traders and are primarily conducted for speculative purposes on CEXs. These transactions tend to be driven by market sentiment and aimed at generating profits from short-term price fluctuations, rather than reflecting the cryptocurrency’s fundamental utility. Thus, our findings are consistent with the evidence in the literature, which argues that speculative drivers dominate the Bitcoin price formation.

Regarding transactions on the blockchain, on-chain demand-side variables tend to be associated with the Bitcoin price in the long-run, whereas both on-chain demand and supply drivers are significant in the short-run. These findings highlight the dual nature of the Bitcoin price dynamics. Bitcoin price movements are related to the speculative behaviour of investors, as suggested by the estimates for the off-chain drivers (in line with the tested hypothesis). In the same time, they are also related to the market fundamentals of the decentralised economy, as indicated by the estimates for the on-chain drivers (which partially contradict the tested hypothesis). These findings suggest distinct strategies of investors trading off-chain compared to those of traders on the blockchain and hence confirm the dual market Bitcoin market.

In terms of macro-financial factors, we find that the price of Bitcoin is primarily related to inflationary pressures and the price of gold in the long-run. Interestingly, inflation and the gold price exhibit a negative relationship in the long-run, suggesting that Bitcoin plays a limited role as a hedge during periods of high inflation and as a store of value. These findings are in line with literature (Liu et al. 2022; Aiello et al. 2023; Kogan et al. 2024), emphasising that investors have developed different valuation models for cryptocurrencies versus traditional assets, such as stocks and gold. In the short-run, however, macro-financial variables are found to be more strongly associated with the Bitcoin price, driven by inflationary pressures and, in particular, stock market developments. This somewhat reinforces the idea that Bitcoin has characteristics associated with a speculative asset, rather than representing a safe haven during periods of high market volatility and inflationary pressures.

As the debate on the development of crypto-assets and their ecosystems continues among the academics, policymakers and investors, many stakeholders are increasingly focusing their attention on the evolving market relevance and maturity of cryptocurrencies, their implications for financial market stability, and the need for a regulatory oversight. The findings of this paper indirectly confirm the presence of an important speculative element in Bitcoin price dynamics, particularly in the short-run. This underlines the importance of strengthening regulatory oversight to enhance the protection of individual investors and minimise potential risks to financial stability, particularly if cryptocurrencies were to play a more prominent role in the global economy in future. Additionally, a greater policy attention may be required to promote financial literacy among investors and to create mechanisms that incentivise the productive use of cryptocurrencies and their underlying blockchain technology, as well as long-term holding over short-term speculation. This would help to mitigate the potential adverse market effects of speculative behaviour in cryptocurrency markets.

When interpreting the results presented in this paper, it is important to consider the conditions and caveats associated with the approach employed in the empirical analysis. First, the ARDL approach used in this study addresses simultaneity bias by incorporating lagged regressors, meaning that the model relies on past values of the dependent and independent variables, rather than their current values to predict the dependent variable. While this mitigates simultaneity bias, it does not fully resolve all potential endogeneity issues, especially when there are strong contemporaneous relationships between the series. Additionally, the ARDL framework does not fully address biases arising from omitted variables or measurement errors, which could still influence the estimated relationships. Since the empirical analysis relies on high-frequency data, the intervals between daily observations are relatively short. Consequently, the lagged values may more closely approximate contemporaneous values, allowing the dynamic relationships between variables to be better captured with a minimal time shift. This reduces the potential for simultaneity bias further. To mitigate remaining concerns about endogeneity, we performed IV regressions as a robustness check. IV estimates broadly confirmed the findings obtained in the main ARDL specification. Second, the ARDL framework is sensitive to the lag length selection. Inappropriate lag orders selection can lead to biased or inefficient estimates, even when diagnostic tests are used for the lag selection. This can affect the model’s overall dynamics and lead to misleading inferences. Third, although the ARDL framework assumes parameter stability over time, real-world data ‒ especially in cryptocurrency markets ‒ may be subject to structural breaks that can affect the robustness and consistency of the estimated relationships. This can bias the long-run coefficients or result in misleading inferences, although short-run coefficients are typically not affected. To mitigate this concern, we explicitly accounted for major structural breaks in Bitcoin prices by including dummy variables in the estimated models. Fourth, it is important to note that the distinction between on-chain and off-chain activities is not always clear-cut. This is particularly true of regulated off-chain fund flows that span both domains, which can result in dual-layered market price formation. Finally, the variables used in our estimations do not directly capture investor motives and behaviour, such as speculation or knowledge of the cryptocurrency ecosystem, which may limit the interpretation of the results.

These limitations highlight promising avenues for the future research. Empirically, the most promising avenues include addressing the identified issues through alternative econometric techniques (e.g. structural break tests or time-varying parameter models), testing the robustness of alternative hypotheses by incorporating different on- and off-chain drivers of the Bitcoin price dynamics (including those that directly capture investor motives and behaviour), and developing a more refined classification of on- and off-chain transactions that can account for hybrid flows. To further improve the validity of inference, a potential simultaneity bias for example between price and exchange-based flows has to be addressed structurally. Conceptually, asset pricing models could be extended to capture the distinct features of crypto investors’ strategies compared to those of traders of traditional assets, such as stocks and gold. Further, different valuation models for on-chain crypto-asset trading versus off-chain crypto-asset trading developed by investors could be formalised.