Cryptocurrency Analysis with Python – Buy and Hold

https://romanorac.github.io/cryptocurrency/analysis/2017/12/25/cryptocurrency-analysis-with-python-part2.html

by HIP_HOP • 9 years ago

89 67 9 years ago

romanorac.github.io

I think you meant HODL.

hendzen• 9 years ago

Don't work with closing prices for statistical analysis. Work with returns (really log returns). This would have simplified some of the code, and made the plots in the same units rather than requiring separate axes.

More detail: https://quantivity.wordpress.com/2011/02/21/why-log-returns/

HIP_HOP6• 9 years ago

Thx... I intend to use it next time. This is going to be a series of blog posts about cryptocurrency analysis.

turnersr• 9 years ago

What do you mean by those plots currently have different units? Don't they consider the price in dollars as the unit?

HIP_HOP6• 9 years ago

They have the dollar as a unit, just y-axis scale is different. For BTC is from $4000 to $20000 and for LTC is from $50 to $350. This is a useful technique to check if curves move together or not.

minimaxir2• 9 years ago

Unless you can include the complete irrationality of the users fueling the crypto bubble to get rich into the statistical model, any statistical analysis of those prices will be misleading.

Traditional historical trend analyses of stock prices assume the market is somewhat rational. (but not 100% efficient, which is where money can be made)

AlexCoventry3• 9 years ago

It's not irrational. It's an asset whose price is rising. If you want to make money, it's a good asset to trade. Just avoid hodling the bag when the music stops...

thisisit2• 9 years ago

Just avoid hodling the bag when the music stops...

and how do you know exactly when the music has stopped?

cortesoft• 9 years ago

If the assumption is that the price will crash, then every dollar made by someone on bitcoin is a dollar lost by someone on bitcoin, since the underlying asset had no intrinsic value at all. Given that there are a lot of people who have more resources than me, and can do better research and better market manipulation, I am pretty sure I have a better chance of being in the group that loses the money.

down2• 9 years ago

if Gold can be irrational for 2k years, I don't see why not bitcoin can't, the only think gold has going for "longer history", yet HN folk love to say "past performance are not an...bla bla", yes it also has the 5% - 10% industrial use, but bitcoin has that too by being able to transfer it without third party, so will also not go to zero because will be useful as transfer of value.

24gttghh2• 9 years ago

Aren't 'miners' the third party?

davewritescode• 9 years ago

Gold has a price floor because it has some industrial utility and is useful for making jewelry.

All that has to happen to destroy bitcoin is a flaw in the implementation.

Seriously, look at OpenSSL and tell me the same issues aren’t possible in bitcoin.

When did we get so trusting of software that we decided it’d be a good idea to base currency on a distributed algorithm?

htormey• 9 years ago

Well ~78% of gold is used for jewelry and only ~19% is used as a store of value:

https://goldprice.com/facts-about-gold

Kinnard• 9 years ago

Are you sure it's complete?

zitterbewegung• 9 years ago

At our local Chicago Python User group chapter (chipy.org) one of the organizers of the Financial Group did a similar analysis. HODL is extremely hard to beat if not impossible.

See https://github.com/Chipy-Finance/CryptoTechnicals/blob/maste...

https://github.com/Chipy-Finance/CryptoTechnicals/blob/maste...

There is more in the github directory also.

HIP_HOP6• 9 years ago

thank you. Gonna look at it

HIP_HOP6• 9 years ago

My next step is to implement strategy testing. I would like to end this series with a classifier that predicts buy or hold.

Follow me on Twitter to get the latest updates: https://twitter.com/romanorac

minimaxir2• 9 years ago

A classifier that flips a coin would probably have higher accuracy.

HighlandSpring• 9 years ago

Give the coin a bias towards long and you've got a winner!

HIP_HOP6• 9 years ago

I am going to try. I am not saying I developed it.

viridian• 9 years ago

This may be a bit pedantic, but the phrase the author uses in his disclaimer "Experts agree that cryptocurrencies are a bubble." bothers me because it's kind of misleading.

The actual fact at hand is ~96% of economists think that the crypto market has been subject to a massive speculation bubble as of November. While that's sort of close to Roman's disclaimer, there is some nuance that ought to be given, since a) 4% of professionals dissent, and b) the question asked of economists had a temporal element to it, and they might have answered differently if the market hadn't tripled in value in as many months. It's the difference between 'the housing market is in a bubble currently', and 'housing is a bubble'.

HIP_HOP6• 9 years ago

I agree, I just wanted to make sure some noob doesn't think he'll invest and make a guaranteed profit.

tbrownaw• 9 years ago

there is some nuance that ought to be given, since a) 4% of professionals dissent, and b)

That's roughly the same % of professionals that I've heard dissent about global warming, yet dissenting about that will get you labeled as stupid and/or corrupt. So what makes it worth calling out for nuance here?

dperfect5• 9 years ago

I ran some tests of my own with simulated automated trading strategies over the past few years of historical Bitcoin price data. I used a genetic algorithm to encode a range of time parameters for MACD, along with different trading triggers and amounts. After a day or two of running it against a huge number of permutations and generations optimized for highest overall return, I came to a similar conclusion: the fittest configurations could do fairly well in automated trading (the more active ones were highly sensitive to trading fees [which were simulated] and market depth [was not simulated]), but over the span of more than a few months, a simple buy and hold strategy produced better returns in almost any time window.

jonny_eh• 9 years ago

Is that another way of saying "the bubble hasn't burst yet"?

dperfect5• 9 years ago

Well, the data did include Bitcoin's first rise above $1000 and subsequent crash (which most people called "a bubble"). If you had bought right before that one "popped", yes - you would have waited a while to be back in positive territory, so that's one time window where you might have lost money compared to the algorithm, but only if you were impatient.

adjkant3• 9 years ago

Depending on the automated strategy, they can dramatically reduce risk though. That has value well beyond the simple return of buy and hold, where if you don't time the sell perfectly, it's all useless. I think that's the big advantage automated trading strategies can have if designed well.

dperfect5• 9 years ago

Yes - I found that to be true in my testing as well. The configurations optimized for highest absolute returns did very poorly in unfavorable market conditions (and max drawdown became significant), while there were some that performed moderately well and were much less affected by downturns in the market. Like most investments, it really comes down to the same thing: risk vs return.

AznHisoka2• 9 years ago

Reduce risk but also involve more time and energy.

thisisit2• 9 years ago

Not to be rude but I stopped reading at "MACD". TA is not the go-to strategy in stock trading for a reason. Indicators like MACD, RSI, Bollinger Bands etc re-paint a lot. So the perfect signals everyone sees on the charts are, for the most part, are just "profitable system" mirage.

dperfect5• 9 years ago

Thanks. The conclusions may be obvious to experts in algo trading (and I heard similar things about the problems of TA for stock trading prior to doing this), but it was educational for me to go through the exercise because (1) it confirmed to me that the crypto markets behave similar to other markets, and (2) it helped me understand exactly where and how a technical indicator like MACD falls short for making trade decisions.

Before trying it myself, I was somewhat skeptical when reading about how these indicators don't work well. The overlays on charts always looked like good buy/sell indicators, so I didn't really understand why they couldn't make for a good algorithm. Only after trying it myself did I gain insight as to why they don't actually work so well.

I guess it's better to learn the hard way than to learn the hard-and-painful way with real money ;)

dmak• 9 years ago

What is an example of a "go-to strategy in stock trading"?

unethical_ban2• 9 years ago

TA is "Technical Analysis", according to Google. You're telling me those indicators are NOT used by rules-based traders in the stock market?

agumonkey• 9 years ago

I find this an impressive fit, did you come up with it on the spot or did you read about strategy simulation somewhere else ?

bhaak• 9 years ago

It's an open secret that "buy and hold" is hard to beat.

This is the reason why ETFs and other unmanaged asset options have become so popular in recent years.

But that insight is not new. The phrase "time in the market beats timing the market" goes way back and also popular stock market participants like Buffet or Kostolany have been saying this for decades.

With regards to cryptocurrencies, we are seeing in a condensed time frame the same development that for example the stock market went through. Cryptocurrencies might become a new asset class but that doesn't mean that it works financially differently than the other asset classes. So it should be no surprise that the same trading strategies end up being the most profitable.

Cryptocurrency Analysis with Python – Buy and Hold

Related Stories