Do the Rich Get Richer? An Empirical Analysis of the Bitcoin Transaction Network

Before Bitcoin, could you ever watch an entire economy's transactions unfold in real time — every payment, every wallet, every dollar moving between strangers? Economists never could. Financial transaction data is among the most guarded information on earth. Therefore, the models built to describe how money flows through economies have mostly been tested against aggregate statistics — wealth distributions and Gini coefficients — rather than the underlying mechanics. You can check that a model's output looks roughly right without ever knowing whether its rules are right. Bitcoin accidentally solved that problem. Every transaction ever made is sitting in a public ledger, timestamped with sending and receiving addresses, and exact amounts. Kondor and colleagues in Budapest decided to look. They modified the open-source Bitcoin client, downloaded the entire blockchain, and extracted a human-readable record of every payment from the system's launch in January 2009 through May 2013. The dataset they assembled is staggering: 235,000 blocks, nearly 17.4 million transactions, and over 13 million distinct addresses. From this, they reconstructed a directed transaction network — nodes are Bitcoin addresses, and a link appears between two nodes whenever money moved between them. The time and amount of every payment came along for the ride, meaning the team could study both the topology of who connects to whom and the dynamics of who accumulates what. To understand how the network is shaped, you first need to know how connected its nodes are. Kondor and colleagues measured degree distributions — basically, how many links each node has — and found that both the in-degree and out-degree distributions follow power laws during the trading phase. The in-degree exponent is approximately 2.18 and the out-degree exponent approximately 2.06. In plain terms, the probability that a given address has k incoming links falls off as k raised to the power of negative 2.18. A small number of addresses have enormous numbers of transaction partners; the vast majority have very few. The network is deeply unequal in its connectivity. The authors also identified two distinct growth phases. The initial phase, lasting until roughly fall 2010, was experimental — activity was sparse, measures fluctuated wildly, and coins were concentrated on just a handful of addresses. Then, from mid-2011 onward, things stabilized into what Kondor and colleagues call the trading phase. Degree distributions settled to stable power laws. The Gini coefficient of the in-degree distribution converged to approximately 0.63, and for out-degree to approximately 0.52. The network also turned out to be disassortative — addresses with many outgoing connections tend to link to addresses with few incoming links — and clustering, while low, sits substantially higher than you'd expect from a random network of comparable size. So the structure is heterogeneous and persistent. The question is why. What mechanism produces a network where a few nodes become massively connected while most remain peripheral? The answer Kondor and colleagues found is preferential attachment. The idea is straightforward: when a new link forms in a network, it's more likely to connect to a node that already has many connections. The rich, in network terms, get richer. This mechanism was famously described by Barabási and Albert and has since been identified in citation networks, the internet, and social graphs. But identifying it in a real financial transaction network — with complete data, not simulations — is new. To test whether preferential attachment actually drives Bitcoin's growth, the authors used a careful event-by-event procedure. For every new link that formed, they recorded the degree distribution at that moment and computed a rank function. If the assumed attachment rule with a given exponent holds, this rank function should be uniformly distributed. They compared the empirical distribution to the uniform distribution across candidate exponents and found the best-fit exponent using the Kolmogorov-Smirnov distance — essentially measuring how far the empirical distribution is from what the rule predicts. The winner is an exponent of approximately one. Linear preferential attachment. The probability that a new transaction connects to an existing address grows in direct proportion to how many incoming links that address already has. Not faster than proportional, not slower. Exactly proportional. That's the network story. Now, follow the money. Here the findings take an unexpected turn. While the network grows under linear preferential attachment, the flow of actual bitcoins through that network follows a different rule — sublinear preferential attachment. The exponent governing wealth accumulation is around 0.8, not 1. Kondor and colleagues frame this precisely: the probability that a node receives new bitcoin is proportional to its current balance raised to the power of 0.8. That difference matters enormously. What does sublinear mean in practice? Wealthier addresses do attract more incoming bitcoin on average. However, the advantage grows more slowly than their current balance. If one address holds twice as much bitcoin as another, it receives less than twice the incoming flow. The marginal benefit of additional wealth diminishes as you get richer. The rich-get-richer mechanism is present but moderated. The numbers bear this out in striking fashion. Kondor and colleagues found that 6.28 percent of addresses hold 93.72 percent of all bitcoin. The Gini coefficient of the wealth distribution stabilizes around 0.985 — essentially as unequal as a distribution can be while still not being one person holding everything. The wealth distribution overall is best described by a stretched exponential, with the extreme tail above 50 bitcoin fitting a power law with an exponent of around negative 1.98. There's also a structural link between how connected an address is and how wealthy it is. The average balance scales with in-degree approximately as balance proportional to in-degree raised to the power of 0.617, across degrees from 1 to 3,000. More connected addresses hold more bitcoin — but the relationship follows a specific sublinear mathematical form rather than a straight proportionality. Month-to-month growth analysis confirms the same pattern: starting balance and subsequent increase are positively correlated, with a fitted slope of approximately 0.857. Wealth grows, but with diminishing returns on existing wealth. The core surprise is this split. Two processes, operating on the same network, follow different rules. Linear attachment for connections — you gain new transaction partners in direct proportion to how many you already have. Sublinear attachment for money — you accumulate bitcoin roughly proportional to your current balance raised to the power of 0.8. The network structure and the financial dynamics are coupled but not identical, and that difference is what keeps the wealth distribution from becoming even more extreme than it already is. What should we make of all this? Bitcoin was designed partly as an alternative to traditional financial systems — decentralized, open, and without a central authority controlling who can participate. Yet, Kondor and colleagues show that the same preferential attachment dynamics that produce hierarchies in citation networks and the internet are at work here too. The inequality that emerges isn't a bug introduced by any particular policy or institution. It's a structural outcome of how networks grow. The broader contribution is methodological as much as empirical. For the first time, econophysics growth models could be tested against a complete, real transaction history rather than presumed random networks validated only at the aggregate level. The microscopic mechanics — which node gets which new link, how individual balances change month to month — were all directly observable. The models aren't being checked against a summary statistic that could be produced by many different underlying processes. They're being checked against the actual process, edge by edge and payment by payment. The implication is uncomfortable and clear. Inequality in a digital currency system may be structurally baked in — not as a consequence of who controls the rules, but as a consequence of how networks grow. Preferential attachment is, in a sense, the gravity of connection: the more you have, the more you attract. Bitcoin's public ledger let us see that force at work for the first time, operating simultaneously on the shape of the network and the distribution of the wealth flowing through it. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.