How to generate trading strategies:
1) Data mining
2) Observation
Observation is simple. You observe and read charts bar by bar for a long period of time. Write down some patterns that crop up.
Test the pattern, if it has good risk-reward metrics, trade it.
1) Data mining
2) Observation
Observation is simple. You observe and read charts bar by bar for a long period of time. Write down some patterns that crop up.
Test the pattern, if it has good risk-reward metrics, trade it.
Data mining is a bit more complicated.
Come up with different factors that could affect the markets.
It could be
- Seasonality (day of week, week of month, month of year)
- price based factors (highs, lows - combined with seasonality - previous day, previous week/mo/y)
Come up with different factors that could affect the markets.
It could be
- Seasonality (day of week, week of month, month of year)
- price based factors (highs, lows - combined with seasonality - previous day, previous week/mo/y)
- Time based factors (enter upon x mins after open, exit x mins before close, enter on x day, exit 1/2/3/4 days/weeks/months later)
- Volume based factors (decreasing volume, increasing volume, > x * average volume, etc.)
- Volatility (increasing vix, decreasing vix, etc)
- Volume based factors (decreasing volume, increasing volume, > x * average volume, etc.)
- Volatility (increasing vix, decreasing vix, etc)
If you're trading options, you can derive factors such as
- Trading ATM, ATM+x, ATM-x strikes
- Buying vs selling option of the specific strikes
and you can get creative with greeks also.
- Indicators - different indicators you think you can use.
- Trading ATM, ATM+x, ATM-x strikes
- Buying vs selling option of the specific strikes
and you can get creative with greeks also.
- Indicators - different indicators you think you can use.
Once you list down such categories of factors that could play a significant role, you then combine 2-3 such factors (don't go more than 4) and generate combinations.
Based on how many factors you list, the combination list would reach 2-3 million or more.
Based on how many factors you list, the combination list would reach 2-3 million or more.
Now, you have to backtest those 2-3 million combinations, and rank them based on few parameters. Risk-reward, CAGR, max drawdown, number of trades, win rate, etc.
Check for statistical significance using statistical tests.
Check for statistical significance using statistical tests.
Once you do that, trade the top 10-25 of the ranked strategies.
This is how few quantitative funds generate strategies to trade.
They
This is how few quantitative funds generate strategies to trade.
They
So, the ideas based on OBSERVATIONS usually have much higher longevity when compared to the ideas that come out due to DATA MINING.
Observation is hard to make.
Data mining is easier to do with computational power.
Observation is hard to make.
Data mining is easier to do with computational power.
That is also why these data mined strategies may or may not work in live, and usually stop working at some point abruptly, and that is also why you have to keep churning strategies to trade, and consistently rank them.
It's almost like a rebalancing a portfolio of strategies.
It's almost like a rebalancing a portfolio of strategies.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
