Crypto Trading 2018 in Review: 17 Advanced + 15 Neural Net strategies tested [Part 7]

Deandree Blake
Jan 8, 2019 · 9 min read

This is Part 7 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 backtests 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

One tough year for Crypto has ended, so I thought it would be a good idea to sum up the year (thanks mark.sch for the idea) and review how much potential it had for trading gains. Even though most coins are down at least 80%, we still had some nice up and down swings with 10s and even 100s of % up in a short time, like April when EOS surged 300+% in few weeks, so I was interested to see how strategies coped with such roller coaster.

Everyone felt how bad this year was for Crypto trading, but I wanted to put it into numbers. Here is chart that shows each coin profits/losses in % (market value) throughout the year, month by month. Bars represent percent change from months first date to last.

I think this gives a good overall picture. While a lot of coins started to drop at the end of 2017, a few still had some momentum that kept going in January. February was kinda OK, but March was one of the worst. April was huge, triple digit gains in 2 weeks, people started counting how many millions single BTC will be worth by the end of the year and how EOS will be Number 1 soon and overtake all other coins, but May brought us back to reality. June kept the same downward momentum, July brought small hope, but after that it was all down, until November, when it was way way down. December brought some recovery, but I think that was mainly due to how huge the drop was.

Testing format

Before we get to the results, few things about my testing format:

  1. I’m still using the same TOP20 coins from previous parts minus VEN and BCH (forked, too much struggle to re-import/test) which makes it 18 coins.
  2. Testing dates are from 1st Jan 2018 00:00 (minus warmup period, calculated separately for each strategy) to 1st Jan 2019 00:00.
  3. Since some coins (like EOS, ADA, IOT, etc) don’t have exchange data for the first few months, I will use first available date as daterange.from.
  4. I’ll test 32 strategies in total. 10 Advanced from Part 2. 15 Neural Net from Part 4. And 7 new, which belong in Advanced group.
  5. For some strategies, I will no longer use default settings. Most of publicly available strategies were made for BULL market and optimized for end of 2017 / start of 2018. Also, they were meant for lower candle sizes (< 60m) which no longer seem be the best time frame (at least in my tests). Before this post, I backtested with default settings and results were quite bad — mostly because way too much trades (multiple per day) or not enough (< 10 over a year). So I made a few optimizations to better adjust to current market and tried not to overfit.

7 New strategies

Most of new strategies use standard indicators, so I won’t dive into mechanics. I tried to find original source, but it’s sometimes hard. All credit goes to authors.

  • SchaffTrendCycle — found here
  • stratego_smaxv7_SL — found in #share-strategies (Gekko Discord Channel) from user xteejx
  • BBRSI — found here
  • HL found here
  • RBB_ADX_BB — found here
  • ATR_ADX — found here
  • w2 — found here

Results for Advanced strategies

Overall I think results are ok, considering market is down 80%. But I’m a bit surprised by few things.

  • I didn’t expect 30m to show any hope at all, but 2 strategies are actually quite good in it.
  • I was expecting 480m to do much better.
  • I wasn’t expecting that some strategies would be down 100% on all coins in all candle sizes. Even with Genetic Algo optimization I was not able to get them to show any promise. My best guess is either they were meant for BULL market or much smaller candle size.

There are few strategies that look very promising:

  • bestone — looking good on multiple timeframes, especially on smaller ones.
  • BBRSI — very good on 120m, one of the best results overall. But terrible on all others. Could be luck.
  • HL — not the best results, but I like that it gets at least some results in all timeframes.
  • IWannaBeRich — overall results close to 50/50, but also seems to work on all timeframes and I think there is some promise here.
  • RBB-ADX-BB — looks very good on smaller 30m/60m timeframes. One of the best results.
  • Turtle — funny how one of the best results come from one of the oldest (1980s) and most well known strategies that has a book written about it. Especially good on higher timeframes (240m/480m).

My TOP3 from this (mixed order) — Turtle, RBB-ADX-BB and bestone.

Let’s take a look at how some of the best results were achieved.

BBRSI 120m

BBRSI making a lot of trades and making them throughout the year, which is important, because there were some strategies, that got good results mainly because playing the April short hyper-BULL market well and then staying out for the rest of the year. Lots of small, consistent trades.


Relatively small amount of trades, which raises suspicion of over fitting. Also, much more trades at the beginning of the year, which means if we tried this live now, could be completely out of sync with market. Catches some big swings nicely at the start of the year, but completely missed all the big moves after that.

Turtle 480m

Does exactly what the Turtle strategy should do — follows the trend and does it nicely. Probably the most stable and predictable strategy here. I like the exits and that it doesn’t keep holding bag of **** like a lot of mean reverting strategies do. And since last year there was a lot of **** to hold, a simple trend following seems like a good choice.

bestone 60m

This is a good example of strategy making most of the gains early in the year and then doing very little. Although it catches one last swing at the end (December recovery).

Results for Neural Net strategies

Maybe some are surprised by how bad this is, I’m personally not. I’ve been digging heavily into the code of Neural Net strategies, so I’m quite familiar with what’s going on underneath the hood and also what’s not quite right in a lot of them. I’ll probably make a separate post where I explain why they don’t work like they should, but the main reason is that there is very little prediction going on. Most of what they do is react to prices after the fact.

“Predictions” usually come in 3 types:

  • Very close to the price, but few candles behind. Which actually makes them just a line of lagged price.
  • Heavily lagged and rounded, looking like a wave.
  • Up and down spikes, way above/below actual price value. Usually end with negative 99% because each next candle is buy/sell/buy/sell and so on.

Here are some examples for 480m candles. Pay attention to light green line, which is price prediction made by Neural Net.

LSTM_MACD_RSI_V3 and mounirs_esto
n8_v2 and neataptic
neuralnet_zschro and NNv2

The only strategy where I can see any chance of prediction going on is ManuNet (image below). It is also the only one that doesn’t predict price, but up/down movements (I describe differences and more details in Part 4).


As we can see, buy/sell logic is overly simplistic, because it tries to react on each direction change. But if you look at the predictions (red/green points at the bottom, green = long, red = short), I think there is some correlation even before the price action. Not in single points, but in general clusters that they form.


My advice is to look at these results critically. One thing that you should keep in mind is that market was very different at the beginning of the year than it is now. First half was still fueled by blind optimism and crowd following, but after that reality started to settle in — this bear market is real and could keep going for a while. A lot of strategies made the most profit in the first part of the year, especially April. My suggestion is for you to backtest more recent data, focusing on last few (3–6) months.

Also, I specifically didn’t post strategy settings that I used. I heavily encourage you try to optimize and tune the strategies for your personal needs and trading style. If you really want, PM me on Discord and I’ll send them to you.

Let’s hope this market pushes us to learn and become better at trading. So when bulls are back in town, we are ready and know how to take advantage of that. Good luck in 2019!


Coinmonks is a non-profit Crypto educational publication.

Sign up for Crypto Anarchist from Coinmonks

A newsletter that brings you week's best crypto and blockchain stories and trending news directly in your inbox, by

Create a free Medium account to get Crypto Anarchist in your inbox.

Deandree Blake

Written by

FullStack Developer by Day | Trader by Night | MCS | Crypto Enthusiast | Gekko Trading Platform Advisor | Backtesting Automation Expert and Platform Developer



Coinmonks is a non-profit Crypto educational publication. Follow us on Twitter @coinmonks Our other project —

Deandree Blake

Written by

FullStack Developer by Day | Trader by Night | MCS | Crypto Enthusiast | Gekko Trading Platform Advisor | Backtesting Automation Expert and Platform Developer



Coinmonks is a non-profit Crypto educational publication. Follow us on Twitter @coinmonks Our other project —