Back to Community
Trading FAANG With PsychSignal Data V. 0

Facebook
Amazon
Apple
Netflix
Google

GOAL

  • Reduce beta
  • Reduce drawdown
  • Maintain Returns

Neither Positivity Bias nor Negativity Bias is being considered at this point. They will be in the future, when I'm finished studying the subjects more.

This is a proof of concept using just increases in bullish intensity. Most posts in the beginning will involve SMAs EMAs etc. I'll move onto more advanced analysis when I have all the data I want to collect.

Current Performance (Algo:FAANG)
Beta 0.82 : 1.14
Drawdow 20.3 : 20.3
Return 225.19 : 142.8
Volatility 0.21 : 0.21

Loading notebook preview...
Notebook previews are currently unavailable.
11 responses

V0.15
Code has some leftover SMA periods from low beta tests - I'll post those next.

Clone Algorithm
52
Loading...
Backtest from to with initial capital
Total Returns
--
Alpha
--
Beta
--
Sharpe
--
Sortino
--
Max Drawdown
--
Benchmark Returns
--
Volatility
--
Returns 1 Month 3 Month 6 Month 12 Month
Alpha 1 Month 3 Month 6 Month 12 Month
Beta 1 Month 3 Month 6 Month 12 Month
Sharpe 1 Month 3 Month 6 Month 12 Month
Sortino 1 Month 3 Month 6 Month 12 Month
Volatility 1 Month 3 Month 6 Month 12 Month
Max Drawdown 1 Month 3 Month 6 Month 12 Month
# Backtest ID: 5987c25c0e6b03544548ba79
There was a runtime error.

V0.02
Low Beta & Low Drawdown at the cost of returns.

Beta 0.49 : 1.14
Return 99.8 : 142.8
Drawdown 8.8 : 20.31
Volatility 0.17 : 0.21

Clone Algorithm
52
Loading...
Backtest from to with initial capital
Total Returns
--
Alpha
--
Beta
--
Sharpe
--
Sortino
--
Max Drawdown
--
Benchmark Returns
--
Volatility
--
Returns 1 Month 3 Month 6 Month 12 Month
Alpha 1 Month 3 Month 6 Month 12 Month
Beta 1 Month 3 Month 6 Month 12 Month
Sharpe 1 Month 3 Month 6 Month 12 Month
Sortino 1 Month 3 Month 6 Month 12 Month
Volatility 1 Month 3 Month 6 Month 12 Month
Max Drawdown 1 Month 3 Month 6 Month 12 Month
# Backtest ID: 5987712adbab994fa426588a
There was a runtime error.

Great. A gain of over 50 percentage points in real profit vs the original in that other thread which |appeared| higher but was actually only 47% because of the hidden margin previously. This version has no margin. Good example of the advantage to us by paying attention to margin.

Since by default there's no way to know margin is there, have to use some tools to navigate those treacherous waters, and this charting is one way: https://www.quantopian.com/posts/margin Margin, knowing, just fine while accidental and unknowing is a sweet siren song luring sailors onto the rocks to their demise. Ergo, know. Know when margin is lurking under the surface, and smooth sailing.

very interesting approach Jacob

I did not use nether PsychSignal nor randomForestClassifier but Generalized Momentum and used
TAANG ('TLT', 'AAPL', 'AMZN', 'NFLX', 'GOOG') to get stability 0.98 in more than 10 year backtest.

Loading notebook preview...
Notebook previews are currently unavailable.

@vladmir: This doesnt use a RandomForestClassifier and TAANG is not FAANG - So no real comparison can be made here. Can you run that with FAANG from 2013 & 2015 onward? Also, its pretty easy to just pick any random acronym that did well in the last decade and apply any working movement indicator to it and get a positive result.

The purpose is to learn how to use the NEW types data that are available to us today, not indicators that have been around since the beginning of quantitative analysis.

jacob,

It is my rule not to use symbols if they did not exist full market cycle.
F(FB) was replaced by T(TLT) because:
1. FB exist only from May 2012.
2. TLT belongs to other than equity asset class.
3. TLT daily returns has negative correlation to any of AAPL, AMZN, NFLX, GOOG.

its pretty easy to just pick any random acronym that did well in the last decade and apply any working movement indicator to it and get a positive result.

I agree.
This exactly what Trading FAANG With PsychSignal Data V. 0 is doing with new, not proved, data types.

FAANG itself is in a way a result of a momentum indicator, right? Jim Cramer talks about how this handful of tech stocks have a lot of momentum and jibe with the millennial economy, and came up with this acronym. So I guess it makes sense to backtest it from the point in time where he created the acronym and onwards. Or better yet, track all his various acronyms from the date he created them: CANDIES, FANG, FAAA, FAANMG -- I don't know what other ones he has or which ones he's serious about.
https://www.cnbc.com/2016/09/27/jim-cramer-renames-fang-as-faaa-your-key-to-long-term-growth.html
https://www.cnbc.com/2017/06/08/cramer-revisits-a-pre-fang-acronym-to-see-if-those-stocks-kept-growing.html

I do get a sense after the June tech sell-out he's not really touting FANG anymore, though I'm not an avid follower or anything.

But then again, is it even worthwhile pursuing Jim Crame recommendations? Sure, he's been mostly right about FANG (except Google right now, OUCH!), but....
http://www.cbsnews.com/news/a-statistical-look-at-jim-cramers-skill-level/
http://www.marketwatch.com/story/jim-cramer-loses-big-in-this-stock-picking-test-2015-10-22
http://www.marketwatch.com/story/jim-cramer-doesnt-beat-the-market-2016-05-13
http://www.businessinsider.com/henry-blodget-professors-prove-that-cramer-isnt-a-lousy-stockpicker-2009-5?IR=T

Don't forget that backtesting this strategy before Febuary 2013 is using hindsight bias. This algorithm would not pick these stocks before FAANG existed and so that data is irrelevant. If you really wanted a robust system you should change it to look at only the top 5 most popular (how to define FAANG?) tech stocks then apply this algorithm to those only. That way you would see the real performance and see if it would have picked FAANG and what it might be picking next if FAANG dies.

Yeah, I think it's a combination of growth and growth estimates. Keep in mind companies like Amazon are technically not in the "tech" sector, despite being very clearly in the tech space.

I chose FAANG because its a benchmark that many people would recognize. This spawned from an algorithm that I wrote that has returned ~900% in the last 3 years by looking @ ~19 stocks that I've been raving about for the past 5 years. That algorithm is much more complex than these - but I don't like that I've skipped over a lot of more simple methods to come up with some convoluted mess - It's like learning calculus without learning algebra - you could probably do it, but there will be gaps in your understanding.

The end goal is for this algorithm to do what I did when I started trading : I scanned social media (most stock twits), monitored post count (manually - horrible), and verified most bullish claims; after 2 weeks I was up 60%... I was good at investing - I'd been paper trading since 2012 (15) But I knew nothing about trading - and I didnt need to. Then school started back up and I didn't have 12 hours a day to dedicate to what I was doing. I put that off and got into algorithmic trading bc I figured if I could automate my process, I could trade and go to school. Jokes on me though - I spend even more time writing algorithms, looking into new data types, and I manage 10's of thousands of dollars (all within the last 2 months - up ~8%) so I can't just put it off. LOL. I've reached the limit of my mathematical knowledge so while I study what I need to know I figured I'd get started on finally automating the process that got me here.