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 5 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
In last part I backtested 15 Neural Net strategies against September. And results were quite good in my opinion, +8 to +12% in best cases. I know a lot of people don’t consider that good, because they have seen screenshots with hundreds and thousands % of profit, which was achieved against the craziest run in the history of stocks/currencies/etc, which might never happen again.
So I just wanted to take a little time to address that. 10% a month? Pffff, I can do that by throwing coins, send me a real strat they say. The best of the best quant hedge funds in the business achieve 20–30% in a good year while struggling with single digits in a more normal year, yet some of us hope to 10x that somehow? If only these quant funds knew that NN lib you have, right? They don’t even Github probably.
If we take a look at Crypto Funds specifically in 2018, situation is much worse. There aren’t much detailed comparisons, but the ones I found report over 20% LOSS and that was in April, but I think we can guess how much worse it’s right now. Pantera, which is one of the most known Crypto funds, seems to be DOWN over 70% year-to-date. If we look at Crypto Index Funds, it’s over 50% DOWN.
Of course, we retail traders don’t deal in Billions, which means we play a bit different game with different rules, but still, we need to keep it real.
And then there’s my favorite — “I just want 1% per day strategy”. Well, that’s exponential. Let’s say you start with small amount — 1000$. And keep in mind — you re-invest the returns. Here’s how it looks:
- In 1 year you’ll get 1.01³⁶⁵ = 37x (3700%) = 37K. Well done, getting some nice profit for your hard work.
- In 2 years you’ll get 1.01²*³⁶⁵ = 1427x = 1.4M. That Lambo is finally yours. I hope you went with yellow.
- In 3 years you’ll get 1.01³*³⁶⁵= 53939x = 53M. You own boats and planes now. Ballin’ 24/7.
- In 5 years you’ll get 1.01⁵*³⁶⁵= 77002912x = 77B. You own a country now.
- In 10 years … The Universe is yours, congratulations.
So my advice to anyone is to get real with your expectations and sync them with reality. Especially since we are in Bear market right now and have been there for quite a while, except for April.
October
Ok back to real life.
So in this part I want to backtest against October, so we can see if the good results continue or if this was just random at play.
But first, let’s do a quick recap of September vs October Market, just so we can keep results in context of how to market itself was doing.
- Month End % is percent change between candle.close of month start and month end
- Month Min % is percent change between month start and months lowest point
- Month Max % is percent change between month start and months highest point
On average, market did 2X worse (-5 vs -11), which means we should expect worse results for October. Here they are:
As you can see, this is worse than September. Almost everything is negative, except for 480m, where few strategies are up 1–2% on average between all coins:
- zuki_nn
- NN_ADX_RSI
- n8
- n8_v2_BB_RSI_SL
- mounirs-ga-version-1
On 360m, only NN_ADX_RSI is positive (1.41% avg)
Everything else (smaller candles) is negative, except for one outlier — mounirs_esto:
- 2.65% on 30m candles
- 1.59% on 120m candles
More results
I know I said (at the end of Part 4) that this time I’ll show prediction lines, but I decided to go in a different direction after seeing results of October. I’ll leave predictions for next part, and instead, I’ll backtest 2 other months. I will choose them based on how the Market did that specific month, and take one relatively bad and one good. It just so happens that these are 2 previous months — July (good, 2%+ on average, with huge spike in mid-month) and August (very bad, -23% on average).
August
Let’s start with the bad month.
As expected, almost everything is heavily down. Except for few interesting outliers:
- n8_v2 on 480m candles — 5.27%
- mounirs_esto on 120m candles — 3.81%
- mounirs-ga-version-2 on 30m candles — negative 1.51%. With small candles and reasonable amount of trades in this month — I’d say it’s surprising and worth attention.
- NN_ADX_RSI on 60m candles — negative 2.38%. Same story, small candles and good amount of trades.
July
Now the good month.
Looks quite good. Best results are on 240m candles and around 7–9% on average:
- Luke_NN — 9.67%
- zuki_nn — 7.89%
- neuralnet_SirTificate — 7.59%
On 360m, best results are in 4–5% zone:
- zuki_nn — 5.73%
- mounirs-ga-version-1 — 4.51%
- Luke_NN — 3.49%
On 480m it’s 5–7%, better than 360m but not as good as 240m:
- mounirs-ga-version-1 — 7.11%
- Luke_NN — 6.30%
- zuki_nn — 5.02%
Summary
When looking at these results, we always need to keep in mind the context — how did the market did in that time period. Also, you need to remember all strategies are tested with default settings, there is definitely room for optimization, hopefully without overfitting. With these Neural Net strategies and with current market conditions, it seems like our aim should be 10+% up in good months and keeping the losses close to 0% on bad months.
Up next — November backtests and prediction lines.
