A bit off-topic, but do you mind expanding a bit on that (linking to the said papers for instance)? I've always though that would hardly be profitable but would love to read up on it.
Incidentally I second the liquidity argument.
Basically whether automated trading can be reliably profitable comes down to whether markets are predictable or not. Bitcoin, Litecoin etc. are rather a special case, but in generalised terms of all traded assets (stocks, bonds etc.), the "Efficient-market hypothesis" is what discusses this, and you should read around. It's a theory that markets are informationally efficient and prices already reflect all publicly available information - if correct, this means it's not possible to consistently make returns higher than the market average. This is far from being proven however, many papers have been published both in favour of and against the EMH.
A nice high level article on a similar topic is http://www.forbes.com/sites/rickferri/2012/03/12/why-smart-people-fail-to-beat-the-market/
- the linked paper http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1356021
on luck vs. skill in fund returns is also very interesting, if not particularly conclusive.
For the Bitcoin market specifically, one thing we can be fairly certain of is that its value is influenced by external events outside the market itself. Some examples include: -
- First major MtGox DDOS (steep fall)
- Worry over the March 2013 block chain fork (fall)
- Cyprus financial crisis (rise)
That helps lend weight to the case for Bitcoin not being easily predictable. Although some advanced stock market trading algorithms do try to factor in that kind of data (look up "sentiment analysis"), that's not really something that cryptocurrency trading bots appear to be capable of at this stage, at least not the ones that I'm aware of!
Being a practical chap, instead of endlessly speculating on issues that haven't been proven (and perhaps can't be proven) either way, it seemed wise to look at what trading algorithms were actually in widespread use among the cryptocurrency trading bots. By far the most prolific are EMA (exponential moving average) crossover based algorithms. These are used by some popular Bitcoin trading bots such as Butter Bot and Gekko. These work by periodically taking a moving average over a short period and a longer period - when these two averages cross, the bot assumes the direction of travel of the price chart is changing, and will buy or sell accordingly.
I did a lot of automated testing of thousands of different EMA algorithms, with a wide range of parameters, using real historical data, as well as various hypothetical scenarios. Like many trading algorithms, they tend to do well in a rising market, and badly in a falling market. Many investors wrongly assume they will protect against losses, but my testing with realistic data proved this not to be the case. Assets don't tend to fall in price in a straight line at a fixed rate (which EMA crossover algorithms would help with) - they bounce around with significant noise, there are panic sells and rallies etc.
A lot of the people using such algorithms have also done historical testing, and come to the different conclusion that they work fantastically well. The problem with that approach is that the Bitcoin market has been almost constantly rising in the longer term. If you'd invested $1000 in Bitcoins a year ago and just hung onto it, Bitcoin would have started at $13 or so, and now be $770, so you'd have made about $58,000, without trading at all. This market average is the benchmark that trading algorithms really need to be compared to. An algorithm that just randomly decided whether to buy or sell every day would still have made a massive profit over that period.
One thing I found very enlightening was to test algorithms over the real Bitcoin price movements, but with time reversed. This creates a hypothetical scenario where the market starts high and ends low, but with hopefully fairly realistic movements and random noise in the progression. I suggest that anyone who is convinced that their awesome EMA trading bot will never lose them money should do this kind of "worst case" testing for themselves and draw their own conclusions (although actually this just represents a fairly bad case and is nowhere near the worst case for an EMA algorithm).
The bottom line is that the current trading bots I've studied can't reliably make money or protect investments when conditions start to turn for the worse. They're heavily dependant on an endlessly rising market. In many cases just buying and holding coins would have beaten them. Whether or not it's theoretically possible for a trader to reliably make profits ahead of the market is a much wider and more open question, which so far hasn't been conclusively answered.