Check out my latest app — TradingGYM. It’s a trading simulator that helps you practice trading with a much faster feedback/learning loop and minimal lookahead bias. Once you open the app, you will see a chart with a random asset at a random point in time. Make trades and fast-forward time to see how it played out.
This is Part 9 in multi-part series:
- Part 1: Basic strategies, introduction, setup and testing vs June-July market.
- Part 2: Advanced strategies and where to find them, testing vs June-July market.
- Part 3: Basic and Advanced strategies testing vs August market.
- Part 4: Neural Network strategies description and backtests against September market.
- Part 5: Neural Network strategies backtest against October market.
- Part 6: Did Neural Network strategies predict November 14th price drop?
- Part 7: Crypto Trading 2018 in Review: 17 Advanced + 15 Neural Net strategies tested
- Part 8: Intro to Statistical Arbitrage in Crypto — Pairs Trading
- [NEW] Part 9: Crypto Trading 2019 Half Year Review: 17 Advanced + 15 Neural Net strategies tested
It’s been a while since my last Gekko strategy comparison, so I thought it’s time for an update. We’ve experienced a very interesting 6+ months, where most of the coins surged really well until the end of June. I hope you’ve been cashing in on this opportunity.
Another reason this has been an interesting period of time is that we’ve seen some unexpected things or things we’ve not seen before:
- BTC dominance going up even more, while most altcoins not following.
- XRP living its own life, showing little to no action at all.
- EOS, while starting strong as always, started collapsing way before other coins and ended up falling heavily through the ranks, currently sitting at No8 by Market Cap, despite being comfortably at No5 spot for a long long time.
- Some unexpected underdogs climbing the ranks, like LTC and BCH.
Raw CSV Results and Configs
For the first time, I’ll be sharing raw results in .csv format and also all strategy configs that I used. They can be found in my Github.
Testing format
It’s been almost a year since I gathered my test coin list, so it was time to freshen it up with more recent top coins. I tried to collect top coins from https://coinmarketcap.com based on Market Cap. Most coins were available, but some were added only recently, so I could not import enough history. I ended up with these 30 coins:
BTC | ETH | XRP | LTC | BCH | BNB | EOS | BSV | XLM | ADA TRX | XMR | DASH | XTZ | NEO | LINK | IOT | ETC | MKR | ONT ZEC | BTG | QTUM | OMG | ZRX | ICX | REP | XVG | GNO | QSH
Market Overview
To get a reference point for results, let’s take a look at how TOP 30 coin prices have changed from Jan 1st to Jul 1st. For better visibility, I have split results into 3 parts — 10 coins each.
BNB leading the way with huge gains, as always. But LTC sudden rise comes as a surprise. Both XRP and XLM struggling heavily and basically not joining other coins in this half-year rally. Also, BTC being one of the top gainers is a bit unexpected.
LINK making huge gains while IOT and MKR at the bottom, relatively moderate gains for other coins.
REP with an interesting pattern — jumping much faster than the rest. Overall the least gains are in this (21–30) range, some very close to 0 % change.
Test Setup
I will be using the same 17 Advanced + 15 Neural Net strategies I’ve been using in previous parts. If you want to read the description of strategies, it’s here:
I’ve also added baseline Buy&Hold (named #BASE in charts) as reference. If strategy has 300% profits while coin is up 500%, that’s not good, and we need to know.
Testing data range will be from Jan 1st to Jul 1st (plus warmup history necessary for each strategy)
Following candle sizes will be used:
30m | 60m | 120m | 240m | 480m | 1440mResults for Advanced strategies
NOTE: I have scaled results in most charts to 500% max. While there are a few outliers (you can check raw .csv if you are interested), 99% of the results fall into this range and the outliers were breaking the scale and making it more difficult to compare different charts, so max 500% was a good threshold to draw the line.
NOTE 2: Due to Medium’s image click-to-zoom feature being temporarily disabled, I had to sidestep the issue by sharing raw full-scale images, they can be found in my Github under /imgs. I’ve also shared more direct links throughout the post.
Now let’s get to the results. As always, Advanced and Neural Net in separate charts, split by different candle sizes.
As expected in such a Bull market, results are good and mostly positive.
A lot of crazy positives come from 2 heavy outlier coins:
- BNB, which was 500+% up at the highest point (vs Jan 1st)
- LINK, which was 700+% up at the highest point (vs Jan 1st)
Although you can see the general trends, like best strategies and candle sizes, the above charts a bit difficult to analyze, so I’ll draw you another one with average profits grouped by candle size:
This gives us a better overview.
First of all — nothing really beats #BASE (Buy&Hold), except on some specific coins. Which is ok, because we are comparing with a period, where there was a clear uptrend and simply holding was obviously close to optimal choice.
120/240/480m looking the best with few strategies leading the way:
- RSI_BULL_BEAR_ADX
- Supertrend
- NEO (best at 60m)
- EMADIV (best at 30m)
Surprisingly, there are 2 strong candidates at 30m — EMADIV and RSI_WR
As usual, let’s take a closer look at some of the best results. To get a better perspective, I’ll try to use:
- BTC as reference.
- Another top coin.
- One crazy result.
NEO @ 60m
Overall I’d say too much trades. XVG case doesn’t look realistic, too many gaps in candles due to liquidity. ETH case looks good, but BTC — worse than you would want.
EMADIV @ 30m
Same — too much trades, subpar results with BTC, good with the rest.
RSI_BULL_BEAR_ADX @ 480m
Good amount of trades, reasonably well on BTC (compared to others), not really catching the big end wave on BNB, partially catching on LINK.
Supertrend
What’s very interesting here — while strategy catches the trend nicely for most of TOP 10 coins, it misses the mark on lower-ranking altcoins, therefore completely missing out on crazy (most probably unrealistic) gains, at least on paper. Overall, I like the stability of this strategy very much.
Results for Neural Net strategies
Neural Net grouped by candle size
For a bit better overview, let’s take a look at the summary with results grouped by candle size.
As usual, 120/240m looks the best, especially 120m.
But, compared to #BASE, all strategies are lacking. Closest ones are:
- LSTM_MACD_RSI_V3 (very close actually)
- neuralnet_zschro
- NN_ADX_RSI
Let’s take a closer look at those 3.
LSTM_MACD_RSI_V3 @ 480m
Upon closer inspection of this strategy, it becomes clear, why results are so close to #BASE. Most of the results contain single purchase somewhere in January, and no more trades. Couldn’t display prediction lines here, because they are so much off that they break the scale of the image. Anyway, I consider this nonworking. I’ll be excluding this strategy from further analysis.
neuralnet_zschro @ 120m
While the overall results are good, what’s happening here is that strategy focuses on mean reversion, while most of the coins are trending. Considering most coins have been trending heavily, it’s just the wrong strategy for this moment in the market.
NN_ADX_RSI @ 240m
While making some nice trades and not many in count (which is good), it’s completely missing some main trends and also predictions are weird.
Luke_NN @ 30m
Too much trades. Focuses on mean reversion and misses the trends. Again — wrong strategy at this moment, except for some coins.
Advanced vs Neural Net
To summarize, let’s put TOP 3 from each category side by side on the same chart to get a better view with a single scale.
Conclusion
Although overall the results look good, compared to #BASE (Buy&Hold) it’s clear strategies are NOT outperforming the simplest strategy of all.
I have to say this is probably not the best time period to test and draw conclusions, because, with such clear uptrend for most coins tested, Buy&Hold is obviously quite an optimal strategy, therefore hard to beat.
Also, some of the best results clearly fall apart once inspected closely, like LSTM_MACD_RSI_V3.
Up next in part 10: I have gathered a few new, yet untested strategies. I will test them against the same time period — Jan 1st to Jul 1st.
Thanks for reading! I hope the Bulls are here to stay.
