
Do Flows Onto Crypto Exchanges Predict Bitcoin Price Moves?
By Bilal Hafeez and Dalvir Mandara
(12 min read)
By Bilal Hafeez and Dalvir Mandara
(12 min read)
Trading View (next 2-4 weeks): We like to be bearish bitcoin.
Investment View (next 1-3 years): We like to be long bitcoin.
This article is only available to Macro Hive subscribers. Sign-up to receive world-class macro analysis with a daily curated newsletter, podcast, original content from award-winning researchers, cross market strategy, equity insights, trade ideas, crypto flow frameworks, academic paper summaries, explanation and analysis of market-moving events, community investor chat room, and more.
Trading View (next 2-4 weeks): We like to be bearish bitcoin.
Investment View (next 1-3 years): We like to be long bitcoin.
In the next few instalments of our weekly bitcoin/ethereum updates, we will empirically investigate several on-chain and flow metrics to learn what relationship, if any, they have with crypto returns. This week, we look at the balance held on exchanges. The balance is simply the total amount of coins currently held on exchange addresses across the platforms we track, such as Crypto.com or Coinbase.
What does it mean if there are large or prolonged periods of outflows of coins from exchanges? One interpretation is investors are not inclined to hold their coins in a liquid capacity on an exchange but instead prefer to keep them in cold storage. This interpretation is bullish given it makes the coins harder to sell and favours the HODLing narrative. That makes sense conceptually, but what does the data tell us?
The daily fluctuations in exchange balance provide some short-term views, but the signals are noisy. A better method is to look at the 30-day change over time – a less noisy signal.
We run simple regressions of 30-day forward returns for bitcoin on the 30-day change in exchange balance (standardised). This should help reveal any relationship between exchange flows and returns – i.e., how much the former impacts the latter. We also want to see how this relationship is changing over time. So, starting from 2011, we progressively add more data to our regressions. For example, the coefficient estimated for 2011 uses data from 2011 only; for 2012, it uses data from 2011 to 2012; for 2018, it uses data from 2011 to 2018, and so on. We find the coefficients are negative and significant (considering p values, etc.) (Chart 2). This supports the view that exchange outflows correspond to a bullish sentiment for bitcoin (positive returns).
The coefficient on the 30-day change in exchange balance has become more negative recently. Notably, since 2017, when bitcoin became more mainstream, the relationship with exchange balance has strengthened versus historic levels.
So, a relationship seems to exist between exchange flows and bitcoin returns. But does the picture change when we look at the same metric across individual exchanges instead of altogether?
We repeat the regression above, this time using the 30-day change in exchange balance (standardised) for each exchange that we track as predictors of the 30-day forward log returns for bitcoin. The other difference is that this regression only uses data from April 2019 onward as that is the earliest date we have exchange balance data for the crypto.com exchange.
We find that the relationship between exchange balance and bitcoin returns differs across exchanges. The coefficient is positive for some of the exchanges and negative for others (Chart 3). That may seem counterintuitive, but it indicates that the relationship is more nuanced at the exchange level. The most positive coefficient is for the Gemini exchange, and the most negative is for Okex.
These simple regressions help provide an empirical view of how (changes in) the exchange balance are related to bitcoin returns. At a high level, the findings indicate that a rise (fall) in aggregate exchange balance indicates a fall (rise) in bitcoin performance. In future instalments, we will also leverage machine learning techniques to better understand non-linear relationships that various on-chain/flow metrics have with crypto performance.
One metric gives a bullish signal this week:
Two metrics give bearish signals:
Lastly, the remaining three metrics give neutral signals:
On balance, on-chain/flow metrics are giving a slightly bearish signal. Here are the details of each metric (with explanations in the Appendix).
Our preferred metric to track institutional demand is flows into bitcoin ETFs. We have seen a return to ETF inflows that has been sustained over the past two weeks (Chart 4). This is bullish bitcoin.
In the short term, a bias for outflows from exchanges exists. Net 42,000 and 30,000 coins have exited exchanges over the past seven and 14 days, respectively (Chart 5). This is bullish bitcoin.
Longer term, the 30-day change in the exchange balance reveals fluctuations in the supply held on exchanges month on month. This metric has been falling since peaking at 72,000 on 1 June, but it remains in positive territory (Chart 6). This means the exchange balance is still increasing relative to the last 30 days, which is bearish for bitcoin.
Together, we view these metrics as neutral for bitcoin in the short term.
Futures open interest has been relatively flat over the past week – it is currently $13.4bn, up just 1% over the past seven days (Chart 7). Around $9bn (68%) of this comes from perpetual futures contracts. Extending to the last 30 days, we find open interest is down 6%. This is bearish bitcoin.
Perpetual funding rates reveal the directional bias of investors. There has been a bias for positive funding rates in general recently. However, on average, they have resumed a downtrend over the past week (Chart 8). This is bearish bitcoin.
The 30-day moving average of the coin days destroyed (CDD) metric ticked up 4pp over the past week (Chart 9). This indicates increased older coin selling. The 1y+ revived supply metric confirms this – it has been in an uptrend since the start of May (Chart 10). In the past seven days, around 78,000 coins that were unmoved for at least a year re-entered liquid circulation.
The 1y+ vintage of the coin supply continues to dominate – over 65% of the coin supply has been accumulating for at least a year (Chart 11). This vintage has been increasing throughout 2022 and shows no signs of slowing yet.
The increased spending of older coins (1y+) can be viewed as bearish for bitcoin as long-term investors move in to take profits, perhaps anticipating further declines. However, a large majority (over 65%) of the coin supply has not moved in at least a year despite this year’s volatile price action and bearish macro backdrop. There is clearly still a strong conviction to HODL among a significant proportion of the coin supply, which can be viewed as bullish for bitcoin.
Together, we view these HODLer metrics as neutral for bitcoin.
On profitability of the coin supply:
Overall, the reduced profitability of the coin supply and realised losses on chain are bearish for bitcoin.
The hash rate is down 2% over the past seven days (Chart 15). The seven-day moving average of the hash rate has been dropping ever since posting highs of around 228 EH/s on 5 May. Miner revenues have started to tick up slightly – up 0.2% over the past seven days after bottoming out on 24 May (Chart 16). Together, these metrics are neutral for bitcoin.
We have introduced a framework for understanding the flow and microstructure dynamics of bitcoin markets. The six key metrics are:
Perhaps the largest institutional vehicle for bitcoin is the Grayscale Trust, with over $27bn in assets. It invests solely in BTC, and so many investors, notably institutional, who cannot hold BTC directly can get exposure through investing in Grayscale. Consequently, if the trust trades at a premium to BTC prices, it may imply ‘excess’ demand from institutions, but ‘excess’ supply if it trades at a discount. Alternatively, the discount may suggest investors have found other ways to get exposure to BTC, whether through ETFs or directly holding BTC. We therefore focus on how the discount has changed in recent months to gauge investor interest. Alternatively, investors may be using other vehicles to get exposure such as ETFs or holding BTC directly. We put more weight on BTC flows than the Grayscale premium.
Another measure of cryptocurrency bullishness is whether investors are willing to hold it in illiquid form (e.g., a private wallet) or prefer a liquid form (e.g., on an exchange). The former would suggest investors are bullish, as they are comfortable with being unable to sell easily. Conversely, holding it in liquid form would suggest investors are bearish, as they prefer being able to sell easily.
Therefore, large flows onto crypto exchanges would suggest investors want to convert their holdings to a more liquid form, implying more bearishness.
We track the growing market of bitcoin futures. Open interest – the sum of long and short contracts – is a good measure of investor interest.
Perpetual funding rates reveal the directional bias of investors. Exchanges set funding rates to prevent a lasting divergence in the price of the futures contract and the underlying since perpetual contracts have no expiry date so never settle in the traditional sense. Consequently, we can interpret funding rates as the cost of holding bitcoin via perpetual futures. Positive funding rates imply longs pay shorts and vice versa. We use it as a proxy for trader sentiment since a positive funding rate implies traders are paying a premium to keep open long positions.
In our introductory bitcoin flow framework, we explained ‘HODLers’ and ‘HODLing’. HODLing refers to buy-and-hold strategies in the context of bitcoin and other cryptocurrencies. Those who HODL for extended periods are die-hard adherents.
We can categorise HODLers by the length of time they have held BTC. We define long-term or staunch HODLers as those who bought BTC five or more years ago and have held it ever since, medium-term HODLers as those who bought 6-12 months ago, and short-term HODLers as those who bought 3-6 months ago.
The coin days destroyed (CDD) metric is defined as the number of coins in a transaction multiplied by the number of days since the coins were last spent. So, increasing CDD suggests older coins are being spent (more coin days are destroyed) and vice-versa.
The percent supply in profit (PSIP). This tracks the share of circulating BTC supply in profit. That is the percentage of circulating BTC whose current price is higher than when it was last transacted (movement).
Net unrealized profit and loss (NUPL). This is the ratio of unrealised profits over total market capitalisation. While PSIP just focuses on whether BTC coins are in profit or not, the NUPL focuses on the size of profits. So, we could have a situation where the PSIP is low – that is, a low share of supply is in profit – but the NUPL could be high if the size of those profits is large.
Spent output profit ratio (SOPR). While PSIP and NUPL focus on unrealised profits or mark-to-market, this measure focuses on realised profits. SOPR is the realised value of a transaction divided by the value at initiation (or creation) – more simply, price sold divided by price paid. If SOPR is above one, investors in aggregate have realised profits, while below one means they have realised losses. In broad uptrends, SOPR spends a significant amount of time above one, whereas the opposite is true for broad downtrends.
When SOPR is rising, sellers are increasingly realising profits. The opposite is true when it is falling. A price rally with a flatter SOPR trend indicates investors are not yet realising their profits with the rally. The reluctance of investors to sell and realise a profit may be because they believe the price will increase further, which would be bullish. At the same time, more profit taking could precede a correction. Typically, buying as SOPR moves around one during bullish periods has proven to be a profitable strategy.
Computing power is central to the crypto market. Miners use advanced computing hardware to solve complex problems that confirm BTC (and other coins) transactions on the public ledger or blockchain. The miners are rewarded with new coins for their efforts. A measure of the complexity of the problems and so the computing performance required to solve them is the hash rate. The higher this rate, the more computing performance is needed to maintain the blockchain. The rate can fluctuate depending on demand for crypto.
OR
Already have a Macro Hive Prime account? Log in