New case study in how on-chain applications can skew important coin metrics

Takens Theorem
5 min readFeb 7, 2020

This short post illustrates using on-chain data to investigate the source of a seemingly anomalous weekly bump in active addresses using Litecoin (LTC) and Bitcoin Cash (BCH). For these cryptocurrencies, active addresses are highest on Tuesdays by as much as 50% or more. I wanted to know why. Here’s my little gumshoe adventure. The adventure begins with intrigue, but ends in a comical anticlimax. So I’ll try to keep it short.

Active addresses

A prominent coin metric is active addresses. In an interval of time, it represents how many unique wallet addresses engaged on-chain (as inputs or outputs). The metric pertains to scaling, value, use cases, adoption, and so on. It’s simple, powerful. I like it.

Active addresses plotted daily over time can reveal curious rhythms. Some projects show an intuitive trend. Bitcoin (BTC) has a robust weekday-and-weekend rhythm: More wallets are active during the week than during the weekend. Ethereum (ETH) has this same rhythm, but less pronounced. See the summary graph below. Others have shown this, too. See here for charts with excruciating detail, along with this rhythm’s historical emergence (thanks to the great and free Coin Metrics community API for data).

Dash (DASH) doesn’t seem to distinguish among weekdays, but Stellar (XLM) does: On Wednesdays, XLM active addresses spike by 1000%. Why? A quick summoning of Crypto Twitter revealed XLM may have had reward payouts on Wednesdays. Party.

The boosts on Tuesdays, Wednesdays are not noise, they are statistically robust departures from the days around them, even in just 4 recent months of data represented here.

LTC and BCH Tuesdays

LTC and BCH show a bump too, on Tuesdays. I summoned Crypto Twitter. I even pestered some on DM. I had no luck (I’m not fancy). I thought: Must be a gambling game of some sort, or airdrop? Maybe deliberate volume boosting?

A Google search turned up nothing. I wondered if I might be missing some information that would improve the search—some clues about what generates this Tuesday perturbation.

So I grabbed transaction-level data across two Monday-Tuesday-Wednesday sequences, one from December and another from January. Total data included 248,720 transactions involving 1,211,571 input-output address pairs. I ensured I could reproduce this bump in active addresses from the raw transaction-level data. Yup:

Data from individual days, across two recent Monday-Wednesday sequences. Tuesday 50%-100% higher.

I started with an eye to volume boosting, because it leaves clues right on the chain: Are there curious patterns in these transactions revealing automated, unnatural on-chain behavior? Patterns over time will have odd distributions if they are not generated by natural usage (consider fake exchange volume and wash trading). My thought was that there’d be overrepresentation of some wallets or transaction sums on Tuesdays compared to Mondays/Wednesdays.

First clue: a plot of total outputs by wallet counts

A clue emerged quickly. For each wallet I plotted how many transactions it sent (x-axis) against how many unique wallets that wallet output to (y-axis). If active addresses are being manipulated, we’d find highly outlying trends on this plot—one wallet might have an extremely and implausibly high number of unique addresses it sends to.

A huge effect. A bunch of wallets, on Tuesdays, have 2,500 outputs in a single transaction. Indeed, this output count is consistent and shows up on both Tuesdays in December and January. When you wander the block explorer through these wallets, that number happens over and over again: 2,500 outputs, 2,500 outputs, 2,500 outputs.

(Side note: See the other outlying wallet along the x-axis? Wild: thousands upon thousands of transactions and with the same input/output wallet. This is the Blitz Ticker. It can account for as much as 50% of BCH’s transaction count.)

The 17 high rollers tossin’ teaser ‘tosh to small-city populations. Graph visualizes about 25,000 wallets.

Lazy follow-up: Google, armed with “2,500”

Armed with that anomalous number, I consulted Google again. Yield came quick. A subtle post on a Reddit AMA:

Developers of an app had an AMA on Reddit here.

A company called Bitcoin Aliens produces a game that gives away BCH every Tuesday as a kind of “faucet,” in exchange for viewing ads. (Or so it appears, as I don’t plan to try it.) Many thousands of individuals have installed this app on Android devices, and its reviews seem positive and at a scale that seems unlikely to have been astroturfed.

Here’s a link to its Google Play page.

But, what about LTC Tuesdays? Lo:

Free Litecoin on Google Play.

What a boring outcome. But is it? A modest giveaway game might be creating a huge weekly trend in a handy metric, active addresses, and on chains that are worth billions of dollars. It is a testament to the instability of chain metrics even on large projects, and the fundamental need to correct for these trends.

But, isn’t this volume “real”? How is it “fake”? Many readers will fondly remember Satoshi Dice, maybe the first to induce these on-chain questions. So there’s further investigation we could conduct to take these questions seriously. Once a user installs them, do the apps send small transactions regardless of whether a wallet is still in play? What if thousands no longer play, but those wallets are still rewarded, flooding the chain with those outputs? It is a debatable issue, of course.

Anyway, I cannot summon a hot take. So I’ll leave this anticlimax with you. I hope you found the little exercise in on-chain data analysis interesting.

If you found this even moderately interesting, give me a chance and follow me on Twitter: @takenstheorem

--

--

Takens Theorem

Dynamic distributed data displays. Intermittent. Friendly.