Data Analysis on Altcoins

[Erichallig]

Hi, do you know about robocoinadvisor? It is the first Robo Advisor
for cryptocurrency forecasts.Intelligent, fast, and neutral forecasts about Cryptos and Tokens secured in the blockchain by AI engine.
RAC will enable futures contracts on crypto currencies. You can check them here
https://roboadvisorcoin.com/crypto-predictions/

[illiki23]

I will look through these.

I started a little coin analysis project for pump prediction as part of some silly idea for a coin/smart contract which locks trades if a pump is detected..  Not sure how to automatically label the movements as pump/dump or not at the time and doing it by hand was difficult.  

One way we can possibly find and label pump/dumps is just looking for 'pump' and 'dump' word frequencies during that time or after. People tend to discuss the pump/dump after it occurs so we might be able to take advantage of this to collect data and build features.  One of the big goals of this project is to link features to coins be it features over time or in general.

There will be natural spikes followed by corrections which look like pump/dumps but are not because no malicious shilling/pumping occurred.  The additional features might let us see differences as the dynamics of discussion would be different.

Erichallig:  I can guarantee that is not the first Robo advisor for cryptocurrency.  The first robo advisors were probably set up shortly after Bitcoin gained value and likely not for public use.

[french_exponential]

Hello guys,

I am new to this topic and I want to contribute !

I agree with illiki23 that sentiment analysis is double-edged but that it can be very powerful. When I saw Electroneum gaining 40% just after McFee tweeted, I was convinced.

I think that sentiment analysis is a good complement of technical analysis. I’ve discovered the R package ‘movavg’ that is very easy to use. I started with a simple 15 days moving average and it is promising: the moving average is either a support or a resistance line, depending on the trend.

I am currently trying out several moving averages with several time frames and weights to find the best one. It would be very useful to automate this process.

It’s a pleasure to join you,
french_exponential

[wesR]

Its always good to have a French on board , welcome!

I was playing around with the criteria mentioned last time and then got dragged into the messy world of few observations that have such stringent criteria (not many coins feature +- 80% swings in 24h - but we all know they exist intra-day). So my main challenge was to display good or bad coins in the first place.

Here a few plots on my way of doing exactly that, it is still work in progress but feel free to comment on it.


So this is my Altcoins dataset: I do not have hourly data, it is all one observation per day (sometimes missing days for certain altcoins), and the measurement is taken at a specific time (I assume). I then narrowed down my research on altcoins with relevance, that is, coins that are actively traded and have at least 100MIO USD$ market cap at the 300th day of this year (just a cutoff). Then, all coins have to pass a threshold volume of 10^4 (not sure if the unit is also USD$ but think so), and applied the logic to data all the way back to 2013.

From this selected raw data, I created new features in the form of weekday, month, year and a few binned categories (quarter of year, month, age of the coin). I continued with more numeric features, given the open close and intra-high and intra-low values. I call them OC (open close) swing (in %), and HL (high low) swing. When OC is positive, it means the coin ended better off than the day before, and I tagged the day with a new feature called goodday = TRUE. Otherwise, it was a bad day, apparently. Now the main part on the menu: data agglomeration. I split the OCswing into positive and negative values (+- 0), then calculated the median of every coin's OCswing and visualized it. The result is seen below.

I would interpret it as the historical mean volatility of those coins with a high visibility. It is quite telling that the market is rather positive due to two facts: 2017 we saw a lot of coins rising fast, while those cryptos that are around for a while had time to settle on a median value. Although the median drowns out the general trend a little bit, I hope you appreciate that some coins are less volatile than others.


Open-Close Swing Med: https://imgur.com/8xt9R1y
All AltCoins fulfilling the condition of being relevant (market cap, volume). Some are more volatile in their open-close values than others, but the median shows to lie on the plus side for all coins, meaning the market was going into a positive direction overall.

BTC OCswings Line: https://imgur.com/5E8RzGc
Bitcoin's OC-swings over time with a loess regression smoothing line. I intentionally didnt use dates, as the labels would have blown out of proportion.

ETH OCswings Line: https://imgur.com/7yFuqXC
Ethereum's OC-swings over time with a loess regression smoothing line, for comparison. No dates as before, but plan to eventually display them together as some point.

Having looked at those trends for a while I started wondering when those terrible swings happen and if one could get out of their way somehow, so I color-coded the weekday along with the OCswings.


BTC OCswings wday: https://imgur.com/1gMTVtR color-coded weekday OCswings for Bitcoin
ETH OCswings wday: https://imgur.com/e7Vj8IW same for Ethereum

And because such plots are not easy to decipher, I plotted them by counts of occurrence with specific thresholds from -20 (and lower) to +20 (and higher). Only thing I didnt have a nerve of doing was to sort the categories for weekdays, not because I am lazy, but because the order of them would be different for every subset of altcoins I use and therefore a repetitive editing task, so apologies for that. I hope the analysis is useful nevertheless.

BTC OCswings wide: https://imgur.com/HDogBaf
As you can see, over 5 years of data, bitcoin prices were starting to get good on a Sunday and followed by a Monday. The party stopped on Wednesday when the percent increase was cashed into other assets. Many wild up and down swings took place on Thursdays.

ETH OCswings wide: https://imgur.com/2RssD0p
More extreme than bitcoin, and seems like ethereum is a good escape when bitcoin need to be moved to other assets. Especially Wednesday seems to be doing well here which is interesting I think, but also Mondays are a good choice.


Merry Christmas and comments below please  !

[french_exponential]

Hello wesR,

Thank you for your post, I read it carefully and I studied your graphs. The evolution of the coins seem indeed to depend on the day of the week, but I think we should perform an hypothesis test to make sure it is not a coincidence.

You should also consider special days of the the year: something definitely happened this Christmas: it seems that people cashed out just before and others got in after. We should also be aware of special days in other countries/cultures.

Hope my thoughts are useful,
Merry Christmas to you too,
french_exponential

[wesR]

I thought the same, and what you suggest is difficult because the data situation is very scarce. Consider that the data I have is from 5 years, means: 5 data points around Christmas time for Bitcoin, and less for all other coins.

On top of having small sample sizes, the granularity is 1/day, which means one cannot even consider dynamics during a day well ( although with HLswings it is possible to a certain degree, depending when for you the day opens and closes). I think what happened this Christmas was unusual. Here a small visualization:


OCswing Christmas: https://imgur.com/YZWitDP
I visualized BTC over the years around Christmas time. Weekdays are color-coded.

HLswing Christmas: https://imgur.com/H11FoPp
This one shows quite strongly that there was a lot of intra-day variability. If I had the raw data in hours/minutes, my resolution would be better, but I think it illustrates quite well what was going on here.


Enjoy the graphs and days at home!

[wesR]

Regarding Altcoins, here is what I came up so far, let me know if that is useful for you:

https://imgur.com/g6ie8hS a few features I use in my current analysis

Preliminary results of some selected coins which a friend of mine had an interest in and I used for this example:

https://imgur.com/iWbb3SR OCswings
https://imgur.com/Wp5h6r8 HLswings
https://imgur.com/vXngU7h Weekday patterns

OCswing: percent swing between open close
HLswing: percent swing between high and low

Seems that some weekdays show seasonality and some of the coins accumulated better scores on particular days, represented by the median open-close swing recorded. I wouldnt interpret too much into it yet as there are like always good and bad days, but perhaps this could be helpful for those who trade out of habit on certain days only.

As always, post your ideas below so I can focus on interesting analyses, thanks!