Skip to content

Latest commit

 

History

History
80 lines (65 loc) · 5.44 KB

README.md

File metadata and controls

80 lines (65 loc) · 5.44 KB
Raw

Context

  • Goal : Bet on the right crypto currency at the right time.
  • Asset : React when a news is published, as of now they all come from binance website.
  • Current challenges : Create excellent extractors

Architecture

We fetch the news every minute on Binance. If the news is new, we extract all of its information (time, title, content), then run an extractor to find on which symbol to bet on.

Then, for each of these symbols (like BTCUSDT), we buy a certain amount of the base asset, like 500$ of BTC (~0.057 BTC).

We then listen to the price change, once a strategy asserts that it is the right moment to sell, we sell.

Here is a sequence diagram of the process : Kapochamo Invest Diagram

Terminology

  • news : data constructed of:
    • url is the unique identifier of the news
    • time in UTC format representing the creation time of the news. Note than the publishing time is usually ~5minutes after that, but lets use only this time for now.
    • title
    • content the body of the news (raw text)
  • symbol : tradable quote asset against a base asset. For instance BTCUSDT is a symbol representing the evolution of BTC expressed USDT. We then call BTC the base asset and USDT the quote asset. You can find all symbols here.
  • extractor : function that analyze a piece of news and extract all the relevant symbols that should be traded. For instance, the extractor onlyBnb will take any news and return BNBUSDT.
type Extractor = (news: News) => Array<Symbol>
  • strategy : function that listens to asset variation, if the strategy decides it is the right moment to sell shouldSell return true. For instance, a simple strategy could be WaitFor(5) that wait for 5 min before selling.
type Strategy  = {
 name: string;
 init: (time: Date, baseAssetPrice: number) => void;
 shouldSell : (time: Date, baseAssetPrice: number) => boolean;
}

Challenges

Create excellent extractors

Today we have relatively good strategies that sell on the right time. They are highly configurable so it is easy to progress. However, extractors should be business-specific. Current ones are agnostic of the information contained in the news, here how they work :

  • relatedAgainstUsdt : find the assets (like ETH, USDT, BTC...) present in the news title and content and return the symbol quoted one USDT.
  • relatedAgainstBnb : same as relatedAgainstUsdt but against quoted BNB.
  • onlyBnb : return BNBUSDT whatever the news is.

It needs to be smarter and consider business logic such as listing, delisting, launches, outstanding trades...

Available data

To do so, I have prepared 12k+ computed performances of 200 news from Binance. You can find in this repository :

  • news.csv with time, title, url, content
  • performances.csv that if the cross product of 12 strategies and 3 extractors with its performance. Note that an extractor can yield several symbols, hence one strategy run with one extractor can return several performances.

For instance the following lines says that if we ran the strategy highest with the extractor onlyBnb would have make a return on investment of 6.27%, trading fee included.

url strategy symbol performance extractor
https://... highest BNBUSDT 6.273632581267119 onlyBnb

Note than even a return of 0.5% is meaningful. If it is ran on 10 news in a span of a few days, total return would be roughly 5% !

Strategies

  • waitFor15 : wait for a set amount of minute, here 15 minutes.
  • highest : theoretical best performances, if we sell at the perfect time. Can't happen in reality.
  • follower1 : sell when the the asset went down by a certain perentage (here 1%) compare to its highest peak.
  • relativeFollower_L5_S3 : sell when we lost 3% of our gain, or if we lost 5% (called pure loss appetite).
  • charly_S10W15L5 : combinaision of waitFor15 and relativeFollower_L5_S10, we wait and then do a relative follower.

Contribute

Why ?

If this works as expected, we can earn a few per cent on every news on any website. Theoretically, we can earn 0.35% per news without any fancy algorithm. With 5 news per week, 4 weeks per month, that is roughly a 7% return per month. Contribute and get early access to this cryptocurrency investment fund without any fee. It is currently running 💸.

How ?

Get the performance.csv file and find an excellent extractor. You can start by filtering with only strategy=highest and extractor not equal to relatedAgainstBnb.

I also developed a tool to see how one asset changes at a specific date : https://api.kapochamo.com/view For instance this screenshot represents the variation of BNBUSDT of this news. /view screenshot

Contact me

Feel free to contact me at kapochamo[at]gmail.com