Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Predicting Price Changes in Ethereum (2017) [pdf]

cs229.stanford.edu

Post reply on HN
Print view View original

21–30 of 42 posts

Re: Predicting Price Changes in Ethereum (2017) [pdf]

#21

Post by anjc »

adjkant wrote: post #17

I've actually run a trading algorithm off a very similar approach for the past 8-10 months, which yielded about 87% return in that time. Using regression I can still to this day hit about 55-60% accuracy. What's not mentioned in the paper is that accuracy is only a small part of the story. If you're accurate 60% of the time but in the 40% accuracy range you're very wrong, acting on the information is useless. As a re…

Very interesting.

Are you buying and selling, or are you shorting too?

Does this 60% accuracy remain if you change the training/prediction period? I'd guess that it would be more accurate if you train based on days versus seconds.

Re: Predicting Price Changes in Ethereum (2017) [pdf]

#23

Post by adjkant »

anjc wrote: post #21
adjkant wrote: post #17

I've actually run a trading algorithm off a very similar approach for the past 8-10 months, which yielded about 87% return in that time. Using regression I can still to this day hit about 55-60% accuracy. What's not mentioned in the paper is that accuracy is only a small part of the story. If you're accurate 60% of the time but in the 40% accuracy range you're very wrong, acting on the information is useless. As a re…

Very interesting. Are you buying and selling, or are you shorting too? Does this 60% accuracy remain if you change the training/prediction period? I'd guess that it would be more accurate if you train based on days versus seconds.

Just buying and selling, all GDAX maker to avoid any fees which would more or less cancel out even the best model I have. Lack of fees is another unique feature of the space.

The training period I keep to 3 months and haven't really moved much since initially trying things. A month or less and the model is overfit and useless, too long and it's not working with current data.

The prediction/"hold time" changes absolutely make a difference. I was running on litecoin and found about an hour to be the sweet spot.

> days versus seconds

Seconds would be useless because you can't trade that fast - minutes is what I use.

Even if accuracy increases with a hold time over days, the average trade value doesn't go up nearly enough to make up for the trade frequency of the minutes/hours level. Why make 3 trades per week with an average value of .5% when you can make 4 trades a day for an average value of .15%? The compounding of that frequency works wonders, and that 4 trades a day for .15% is what I was actually hitting for a few months.

For the record, I do also compare to both naive buy/hold over the training period and the average trade value for the period for all times (different than the buy/hold time because I have a profit lock-in threshold for individual trades, also tuned with genetic algorithms), and the model outperforms both still.

The model is still predicting positively but the average trade value is shrinking + market thinning hence why it was breaking even recently until shut down.

Re: Predicting Price Changes in Ethereum (2017) [pdf]

#24

Post by adjkant »

blinds wrote: post #22

Applying decades old statistical techniques to predict trends in time series data. Now with added blockchain, as is tradition.

Same old ideas as stock prediction, yes, but the crucial difference here is a market with high volatility and no trade fees. Circumstance is everything.

Re: Predicting Price Changes in Ethereum (2017) [pdf]

#26

Post by tyrex2017 »

nanis wrote: post #14
anjc wrote: post #11

Earlier quoted context omitted.

It looks like it is a class project. But so what? It's giving you enough information to reproduce the study, which is more than can be said for many published research papers. That makes it useful imo.

As a class project where kids are learning how to use software, clean data, type equations etc, it is acceptable. It is unreasonable to expect them to know better when they are just getting a feel for the tools. If this is supposed to give insight to people on how to allocate their money, the critiques would have to be much harsher.

i have read papers in the best journals. not better, not more insightful, not deeper, than this one. in fact, they usually would have included only half the models.

the only question i have: such worse performance by nn is strange, it should be similar to logistic regression..

Re: Predicting Price Changes in Ethereum (2017) [pdf]

#27

Post by mirekrusin »

anjc wrote: post #19
fwdpropaganda wrote: post #18

Earlier quoted context omitted.

> Interesting that even the most naive methods still have >50% accuracy. All methods that you'll ever see have >50% accuracy, because if you find a signal with 50% accuracy. Here's a bit relevant to this conversation: > Previous work on predicting the directionality of Bitcoin prices has shown that significant signal exists in the price of the cryptocurrency. Hegazy and Mumford (2016) compute an exponentially-smoothe…

I presume that the context of those accuracy numbers is prediction for the next period (i.e. 8 minutes, 10 minutes, 1 minute). To me 60% sounds good, but apparently not to people in this thread :)

Random generator has 50% prediction accuracy. For series like ether where price globally/on average was rising, constant predictor of “will rise” would have similar accuracy.

Re: Predicting Price Changes in Ethereum (2017) [pdf]

#28

Post by blinds »

adjkant wrote: post #24
blinds wrote: post #22

Applying decades old statistical techniques to predict trends in time series data. Now with added blockchain, as is tradition.

Same old ideas as stock prediction, yes, but the crucial difference here is a market with high volatility and no trade fees. Circumstance is everything.

Yes, there is a difference in circumstance. Ironically though, their best performing model assumes constant volatility...

Re: Predicting Price Changes in Ethereum (2017) [pdf]

#29

Post by heptathorp »

> A logical explanation for the high volatility of Ether, relative to Bitcoin and common stock, is that Ether is transferred or traded in a way in which Bitcoin and stock are not. More specifically, Ethereum has contract accounts which can cause Ether to be transferred between accounts in an unpredictable manner. (While Bitcoin and common stock are also traded by algorithms, the distinction is that anyone - even someone writing a hello world program - can instantiate a contract account on Ethereum. Meanwhile, algorithmically trading Bitcoin or stock requires more technical sophistication.)

Can anyone clarify what the movement of Ether between contract addresses has to do with the price of Ether? Intuitively, the price of ETH would move when it is traded against other currencies, not when it is transferred between accounts on Ethereum. Yes, it's easy to write a contract that moves Ether. It's also trivial to create a Bitcoin transaction.

Re: Predicting Price Changes in Ethereum (2017) [pdf]

#30

Post by fenwick67 »

mrwebmaster wrote: post #25

Maybe there are a lot of other trading algorithms that worked for some time but ended with a loss so big that exceeded all the profits, and we will never see those stats. Publisher bias.

That's what I think is quaint about all these people who came up with crypto buying strategies. The naive "just buy as much as as you can" strategy was very effective until January.
Post reply on HN