February 22, 2023: The Fiddy Indicator 📈📉
Is it Real? Could you profit?
The “Fiddy Indicator” is catching on! Occasionally we’ll see it cited in various technical analysis channels around the crypto sphere.
For those who failed remedial trigonometry, the “Fiddy Indicator” tracks the balance of Tether ($USDT) in 3pool. Lower amounts of Tether allegedly correlates with higher $ETH prices, and vice versa. The effect is being tracked on the Blockworks Research Dune dashboard.
Is the effect real? Better yet — is the effect predictive?
Today we put it to the test. For the impatient, here’s the TL/DR topline conclusions:
Over a 1.5 year timeframe, we uncovered data supportive of the “Fiddy Indictor”
The “Fiddy Indicator” potentially demonstrated some predictive power over future Ethereum price trends.
Study of more varied time windows needed for further validation.
If you’re looking for financial advice, look elsewhere. This only contains data analysis.
DATA
We wrote a quick script to scrape historical 3pool balances relative to ETH price and executed it via Alchemy using Brownie.
Then we pulled the data into a Jupyter Notebook. Here’s the ETH prices (red) versus the 3pool balance (blue) we have to work with. That blue/red flip is right off the bat bullish for an inverse correlation between the two stats.
Our data covers the timeframe from July 13, 2021 to present (TriCrypto launches on-chain on this date and helpfully serves as our historical price feed). During this window, ETH first mooned to near $5K, then back down to frown town where we woefully sit at present.
DISCUSSION
We start by looking at a correlation table to see just how strong the high level correlation actually is. We consider Ethereum price against a range of other factors.
How you should interpret this chart… a solid color is a very strong correlation — solid blue means they run exactly with each other, solid red means they run perfectly opposite each other. Either way solid means the pattern is strong and useful. The variable we’re tracking is the current price of ETH (“eth_price”), which unsurprisingly correlates 100% with itself for that beefy blue box at the top.
In contrast, a pale white 0 is the worst result — meaning absolutely no correlation. We tossed in a column of random numbers (“rand”) to help visualize what very weak correlation looks like. In this case the ~3% correlation basically rounds to zero.
We tossed in several factors that all correlated to some degree with ETH price. Bitcoin price correlates a whopping 98% with ETH price for the window, basically meaning BTC and ETH are joined at the hip like the Dow Jones and S&P 500.
Let’s not bury the lede though… the Fiddy Indicator displays a meaningful inverse correlation with 3pool Tether balance for this window. For the past year and half, the balance of Tether in 3pool exhibits a -49% correlation with the price of Ethereum. Not perfect, but enough to suggest something may be afoot. It’s a good indication the Fiddy Indicator may be at least partly correct. You can see the effect on a scatterplot.
How substantial is the effect? The roughly ~50% correlation is nice, but also far weaker than the other factors we included in the correlation analysis.
For instance, we also observe a stronger -79% correlation between the current block height (aka time) and the price of ETH. In other words, for the past year and a half the trend of ETH has been sharply “down only.” Similarly, ETH price also correlated more strongly (78%) with the total TVL in 3pool. These are both clearer signals over the timeframe.
We want to untangle any potential confusion among correlations here — ice cream sales go up in summer, and because more people swim, so do drownings — but the “ice creams to drownings” correlation exists but is meaningless. Similarly, we need to test if the Fiddy Indicator is directly correlated with ETH Price, or if it’s confusing correlations with other variables. That is — it may be the case that Tether balance is correlating with 3pool TVL, and we’re mistaking the effect because 3pool TVL better correlates with ETH price.
We can get some insight by looking directly at the correlation analysis for Tether balance against other factors.
The results here work in favor of the Fiddy Indicator. The strongest correlation with the Tether balance in 3pool is Ethereum Price, with every other factor showing far less overlap. It’s a bit remarkable — all these other factors show stronger correlation with ETH price, yet less correlation with Tether balance of 3pool. Even though BTC and ETH run together 98% of the time, there’s a big gulf between the Fiddy indicator correlating with ETH price versus BTC price (~7 percentage points).
So, while other explanatory factors may still exist, all the evidence so far points to the validity of the Fiddy indicator — there does indeed appear to be some special connection between ETH price and 3pool Tether balance.
With this context out of the way, we’ll drop the full correlation table for you, so armchair data scientists can interpret as they like and suggest your own conclusions.
Now the fun part — let’s try to tease out causation not correlation. We know “correlation does not prove causation” trolls emerge around the second grade and think they sound smart. Proving causation is trickier, but not impossible.
You’d probably agree that the 98% correlation between Bitcoin and Ethereum prices are essentially meaningless for the greedy trader. If you observe Bitcoin tanking, this doesn’t give you any edge in trading Ethereum. Both tank at the same time, so you can’t jump to buy ETH when you see BTC trend upwards.
To be able to trade the Fiddy Indicator and make money, you want evidence that a change in the 3pool Tether balance occurs before Ethereum prices move, not simultaneously or after.
To tease out this effect, we refactor our dataframe a little bit to check what we care about. The dependent variable is the next recorded percentage change of the Ethereum price. We want to see if a change in Tether balance today correlates with a change in Ethereum price tomorrow.
For this analysis we’re converting all variables from raw prices/numbers into their percentage change. This helps to normalize the indicator because it’s more useful to observe directional change as opposed to rough aggregate prices. As a trader you care if the next move is down 10% or up 10%. You don’t care to if I can predict that ETH is likely to be “around $1600” tomorrow — that’s trivial and meaningless since the price is around $1600 today. If I can confidently predict the price will go up 10% tomorrow, you’re ecstatic because you’re making money, you’ll pocket a gain whether today’s price is $16 or $16,000.
Predicting “percent change” is the more useful and more challenging metric. For this reason, traders should care more about running correlation analysis on percentage changes, not raw numbers. “Does today’s diretion impact tomorrow’s price direction?”
Most importantly though, we time shift the dependent value, to see if today’s changes propagate into the future. All this lets us reframe our analysis to explore the most relevant question for traders: “If 3pool Tether balance is trending downwards today, will ETH price trend upwards tomorrow?”
To set expectations, we generally expect low signal when looking at such trends. If you found a 100% correlation, you’ve found a crystal ball. You can make money every trade because you can see perfectly into the future. This doesn’t really exist in any meaningful fashion. Quantitative analysts are too busy killing themselves looking for an indicator that works reliably better than 1% of the time. If you can get a predictable 1%, it’s enough of an edge to make profits long run.
With this in mind, let’s also calibrate our expectations by looking at the unshifted data. This is the easy case — within the same block, how do directional trends correlate? Similar analysis to above, but on percentage change instead of absolute number. Here is the correlations for ETH price direction in the same window.
Again, we see similar correlations as above, though the effect is not quite as strong — BTC directionally moves 87% with Ethereum, though their absolute price levels were about 98% correlated here. Their prices stay in a close range, but they don’t always skate in perfect synchronicity.
Similarly, the 49% inverse correlation between 3pool balance and price becomes 21% when we’re just looking at pure directional movement. At 21%, it’s better than random noise, but a rather slight edge.
This is for unshifted though — no alfa in predicting what just happened. By the time BTC is going down, ETH has already nuked with it. Let’s shift the ETH price… to see which of today’s trends affect tomorrow’s prices.
Unfortunately, whenever you try reading into the future, it gets murkier.
Unsurprisingly, while BTC and ETH move together in the same minute, you can’t claim that a move in BTC today means ETH will dump tomorrow. The correlation here is basically zero, even worse than the random variable we added as a sort of control.
Encouragingly though, the correlation between 3pool Tether balance directional change does not completely disappear. An 8-9% inverse correlation suggests a potentially meaningful signal. It’s possible it’s just noise, but at 8-9% I’d want to dive in further and see if it’s real.
Remember, when running this sort of analysis, traders would be thrilled to get a 1-2% edge if it’s meaningful and not noise. Multiply 1% by a million transactions and you make essentially guaranteed money in the long run — this sort of predictable edge is how casinos make bank.
Before you get excited and pursue this too aggressively, note that an 8-9% edge is messy and all but guaranteed. Recall the above scatterplot, which was 50% correlated and demonstrated a pretty clear directional trend.
In contrast, heres’s what our single digit correlation looks like:
For this, one or two outlier points may be badly skewing the results. Prune a few of these black swan events and the edge may disappear entirely. Anybody interested in trading the Fiddy Indicator should run a lot more backtesting our rough cut.
In particular, this cursory analysis has some major holes. Most troublesome, we picked a granularity of 10,000 blocks to scrape our data (roughly 1 day). We couldn’t get much finer than this without the RPC node crashing.
But a day is a lifetime in crypto! A change today causing an impact tomorrow is almost absurd to discern given quite how much happens in 24 hours. Running this analysis again, but testing finer time intervals (ideally a block or handful of blocks) might expose different and potentially more actionable trends.
In conclusion:
Over a 1.5 year timeframe, we uncovered data supportive of the “Fiddy Indictor”
We could not invalidate the null hypothesis, that the “Fiddy Indicator” has no predictive power over future Ethereum price trends.
This is not financial advice, make your own coffee at home.
Stay safe, degens!
Super informative article . I was not aware of this relationship. I cant believe you put out this much Alpha on a near daily basis. Felt like this article alone was worth the subscription. Keep up the good work fren!