Sentiment Analysis


Hi Everyone,

This week markets jumped from $400/BTC to $450/BTC on the announcement that Paypal would be partnering with Bitpay, Coinbase, and GoCoin to allow its merchants to accept bitcoin:  Since then the price has tumbled to $370/BTC.  It further shows the ineffectiveness of fundamental news on price in an environment where short term buying and selling pressure dominate price discovery.

  • Swiss secret service to use bitcoin to pay informers:
  • Last week's rumor that BFL was shut down by regulators was substantiated this week:
  • Gavin Andresen begins work on invertible bloom lookup tables to make Bitcoin more scalable:
  • The supposed attack on the Monero network looks to be a hoax.  A good summary of events:
    • A reputable hacker and pool operator BCX claimed he had discovered a fatal flaw in Monero.
    • No one knows exactly what this flaw is or if it even exists.
    • BCX threatened to attack the network on 9/24.
    • No attack came but BCX now claims it will take a few days for the effects of the attack to manifest.
    • Some people think the original intended attack was related to a long-debunked attack vector (
      • At a high level, a person's public key gives you an equation involving the person's private key and a ring image also gives you an equation involving the person's private key.  But it is not the case that with these "multiple equations" you can solve for the private key.
    • Others suspect this is purposeful FUD by BCX to drive the price down and buy cheap coins.
  • Peercoin loses earlier gains as Nubits launches:
    • The article isn't exactly right in that Nubits has nothing to do with Peercoin (PPC) since it is an separate altcoin since there is some mechanism in the Nubits/NuShares system which pays dividends in PPC.
    • I've only skimmed the arduous whitepaper ( and it seems Nubits is trying to do something similar to BitsharesX's market-pegged assets but instead of using margin calls and market psychology to peg the price, it uses data feeds and centralized interest rate oracles.
    • I'm highly skeptical of all pegging schemes without ultimate redeemability for or delivery of the underlying asset.
    • On a side note, apparently BitsharesX is also using data feeds now to band the price of its bitAssets after their market peg broke down a few weeks ago:  Disregard the title of the linked thread.
  • IOCoin (not to be confused with I0Coin) switched to a new form of PoS where transaction fees are burned instead of given to stakers.
    • So while staking seigniorage provides inflationary pressure, burned transaction fees provide deflationary pressure set by the market (i.e. transaction volume on the network).
    • Even after reading their short, badly-written whitepaper (, I'm not sure what the point of this is.
  • Secret Goldman tapes surface:
It seems like the cryptospace generates copycats of the flavor-of-the-week at an alarming rate.  When anonymous coins started getting popular after Darkcoin's rise, every week saw a new coin based on a new anonymity mechanism, where each successive mechanism was more convoluted and more difficult to decipher than the last.  There was also the hype behind arbitrary asset issuance protocols like Nxt, Counterparty, followed by several others.  Now with Nubits, it seems pegged assets is the new hot thing.  I think, at this point, what's necessary are experts in mechanism design ( and implementation theory ( to come into the cryptospace and separate the plausible from the impossible.  I also feel that many altcoin whitepapers are needlessly long and painful to read.  That being said, I think anyone who has the patience to thoroughly sift through the noise will find many opportunities in the altcoin markets.

In an unrelated note, I recently read a paper on the correlation between bitcoin market activity and Twitter sentiment:

To summarize its findings:
  • The paper looked at the number of times positive emotive words (e.g. happy, great, awesome, etc.), negative emotive words (e.g. sad, bad, unhappy, etc.), and uncertain emotive words (e.g. hope, worry, fear) were mentioned in tweets alongside the word "bitcoin".
  • Correlation between negative tweets (but not positive tweets) and bitcoin price is found to be negative at the 1% statistically significant level.
    • This means negative sentiment is reflected in the market much more than positive sentiment (i.e. it could be that the baseline for tweeting is that most people say positive things about bitcoin most of the time)
  • Correlation between uncertain tweets and bitcoin price is found to be negative at the 1% statistically significant level.
    • This means that price goes down as uncertainty goes up.
  • Trading volume was correlated with positive tweets, negative tweets, the ratio of positive to negative tweets, uncertain tweets, and total emotive tweets at the 1% level.
    • This means that more twitter action goes together with more trading volume.
  • Wider spreads are correlated with negative sentiment.
  • Narrower spreads are correlated with the ratio of positive to negative sentiment.
  • It also found that sentiment (negative, uncertain, and positive to negative ratio) could lead price movements by up to 2 days but sentiment did not lead trading volume.
    • This suggests that directional sentiment lingers for a few days but excitement does not linger.  In other words, people who are right in their sentiment but slow pay off people who are right in their sentiment but fast.
  • Ultimately, the methodology could be better.  It would have been better to use price returns instead of price and de-trended tweet counts instead of tweet counts since the correlation of two non-stationary time series is less telling than the correlation between two stationary time series since correlation is a linear relationship.
  • Also they could improve upon their word choices for what represents positive or negative emotions.  They could also look for not just the word "bitcoin" but its many variants like "BTC" or "XBT".
    • If you want to get really fancy, you can build out a neural network of emotive words each with a weighting for "positiveness", "negativeness", and "uncertainness" off a list of seed words (e.g. happy = (1,0,0) (positive, negative, uncertain), sad = (0,1,0), unsure = (0,0,1)) and doing heavy machine learning as a first pass over Twitter.  Then in the second pass, apply the full neural network onto the word "bitcoin" plus variants.
I think there is a lot to explore on the topic of sentiment analysis in the bitcoin markets.

Kevin & Team Buttercoin
Bitcoin Trading Made Easy | 

Buttercoin currently in Private Beta, if you'd like access right now apply here: