Hi guys, I am totally new to automation trading. Can someone explain the complete process? Start from the scratch please
Rename
> Start from the scratch please
I assume with "automation trading" you mean fully automated trading in contrast to manual trading.
Let's start with a quick analysis of "manual trading".
As a trader you use some information (news, rumors, gut feeling, observation of price action) to arrive at a trading decision.
Such a decision says "Buy or Sell the security XYZ and thus open a new position or close an existing position."
Then you login to your broker account and enter such an order.
It is possible to automate this whole process but on closer inspection there are quite a few things to consider.
I'd like to divide the process in two parts:
1.) Generate a trading decision
2.) Generate an order in your broker's account.
Part 2.) is covered by the post mentioned by cone in the previous post.
Part 1: generating a trading decision
In order to generate a trading decision you need just two ingredients:
a) a set of sources of informations
b) an algorithm (a rule, process, formula) to condense all this information into a single decision: What to buy/sell and when.
Information sources
Because the process ought to be fully automated (i.e. executed by a computer) the incoming information needs to be "machine readable". For the development/test of b. (the algorithm) we need some larger set of historical information. This historical information is used to see if the algorithm worked in the past (and we hope it continues to do so in the future), a process called backtest.
Possible information sources are:
Past prices and price action/movements.
This information is available from a large number of historical data providers, see zhe extensions "Data Extensions", "EODHD", "IQFeed", "Norgate Data" and so forth. There is a Meta-Extension called finantic.BestData which combines several (always error-prone) data feeds into a single, combined feed which has presumably less errors.
Various Indicators based on prices
It is simple to calculate a large set of indicators from the prices, one example is Rate of Change another the big family of moving averages. This discipline is called "technical Analysis", which is a big misnomer because it is neither very technical nor related with any useful analysis.
WL comes with a large number of Indicators and a number of extensions add to this set.
I assume with "automation trading" you mean fully automated trading in contrast to manual trading.
Let's start with a quick analysis of "manual trading".
As a trader you use some information (news, rumors, gut feeling, observation of price action) to arrive at a trading decision.
Such a decision says "Buy or Sell the security XYZ and thus open a new position or close an existing position."
Then you login to your broker account and enter such an order.
It is possible to automate this whole process but on closer inspection there are quite a few things to consider.
I'd like to divide the process in two parts:
1.) Generate a trading decision
2.) Generate an order in your broker's account.
Part 2.) is covered by the post mentioned by cone in the previous post.
Part 1: generating a trading decision
In order to generate a trading decision you need just two ingredients:
a) a set of sources of informations
b) an algorithm (a rule, process, formula) to condense all this information into a single decision: What to buy/sell and when.
Information sources
Because the process ought to be fully automated (i.e. executed by a computer) the incoming information needs to be "machine readable". For the development/test of b. (the algorithm) we need some larger set of historical information. This historical information is used to see if the algorithm worked in the past (and we hope it continues to do so in the future), a process called backtest.
Possible information sources are:
Past prices and price action/movements.
This information is available from a large number of historical data providers, see zhe extensions "Data Extensions", "EODHD", "IQFeed", "Norgate Data" and so forth. There is a Meta-Extension called finantic.BestData which combines several (always error-prone) data feeds into a single, combined feed which has presumably less errors.
Various Indicators based on prices
It is simple to calculate a large set of indicators from the prices, one example is Rate of Change another the big family of moving averages. This discipline is called "technical Analysis", which is a big misnomer because it is neither very technical nor related with any useful analysis.
WL comes with a large number of Indicators and a number of extensions add to this set.
Fundamental Data
There is a large set of fundamental data provided by an extension called the Fundamental Extensions. For example the FRED indicators are a huge set of macroeconomic data. Fundamental data usually changes not faster than once a week, often just once a year and thus is better suited for long term trading strategies.
News
Until recently news was not a machine-readable source of information. This changed radically with the event of ChatGPT in November 2022. ChatGPT (and other, simira Large language Models) is able to "understand" news articles and assign a sentiment value that in turn can be used to base trading decisions on it. To leverage this source of information you need an extension which feeds news articles into WL like the finantic.News extension. Second you need a way to send a news article to a Large Language Model and build an indicator from the returned sentiment values. This is done by the finantic.NLP extension.
Other information sources
Of course there countless other sources of information like "Moon Phase" or simply "Day of Week". To use such sources it is often necessary to use some external library in a coded strategy or even build a specialized extension. If yo are not a coder it is possible to use WL's consierge service to get these kind of task done.
There is a large set of fundamental data provided by an extension called the Fundamental Extensions. For example the FRED indicators are a huge set of macroeconomic data. Fundamental data usually changes not faster than once a week, often just once a year and thus is better suited for long term trading strategies.
News
Until recently news was not a machine-readable source of information. This changed radically with the event of ChatGPT in November 2022. ChatGPT (and other, simira Large language Models) is able to "understand" news articles and assign a sentiment value that in turn can be used to base trading decisions on it. To leverage this source of information you need an extension which feeds news articles into WL like the finantic.News extension. Second you need a way to send a news article to a Large Language Model and build an indicator from the returned sentiment values. This is done by the finantic.NLP extension.
Other information sources
Of course there countless other sources of information like "Moon Phase" or simply "Day of Week". To use such sources it is often necessary to use some external library in a coded strategy or even build a specialized extension. If yo are not a coder it is possible to use WL's consierge service to get these kind of task done.
QUOTE:
The trading algorithm
With all the information mentioned above in place you need a way to extract a trading decision form that very information.
In a completely automated trading process this is done by an algorithm. This is a set of rules and called a "strategy" in WealthLab.
You can get ideas for a trading strategy from journals, books, internet posts, YouTube videos or published strategies on www.wealth-lab.com and so forth.
You'll have to translate these strategies into an executable form, with WL these are Building Blocks or a strategy coded in C#.
The next step is to test the strategy with historical data, this process is called a backtest. It turns out (and I did this many times during the last 20+ years) that most ideas from journals books or the internet do not survive a backtest i.e. these ideas are not profitable if applied to more than one symbol and/or for more than a few weeks of historical data.
In fact it is rather difficult to find a strategy that has a true "edge" i.e. works for a larger portfolio over five or ten years.
Once you found a profitable strategy it might be time to "optimize" the various parameters used in such a strategy. And this far more difficult than most people think. It is very easy to get spectacular results through an naive optimization process. And a strategy that works in the past (exactly for the data range used during optimization) but not in the future - worthless. But I think this topic is for another thread...
Your Response
Post
Edit Post
Login is required