Sorry, I am not familiar with Amibroker, I have developed my own python backtesting engine as it gives me the flexibility to make customized frameworks and add new features as per my needs.
Zerodha provides a very nice python kite connect api package which can be easily used with any python backtester and it streams live data and we can place trades in real time. Also they provide historical data for backtesting.
Try Using a python open sourced packages as it will make our life easy, for this system I have used the python reinforcement learning framework, RL is a supervised machine learning algorithm which can be trained and later it will learn by itself. This algorithm is usually used in training computers to play chess, crawl web, sentiment analysis etc
You can check this article for an overview:
https://neptune.ai/blog/best-reinforcement-learning-tutorials-examples-projects-and-courses
Example trading scenario:
https://towardsdatascience.com/deep-reinforcement-learning-for-automated-stock-trading-f1dad0126a02
I will just give a hint on how I have built this system, first just look at charts and analyze how you would need to trade it manually:
Preprocessing steps:
1. Select the time frame to trade say 5 min.
2. Identify all the candle pivots (3 candles, with lower highs or higher lows on either side)
3. Mark the structure for all the identified pivots.
4. If needed you can also use some technical signals.
Analysis rules and training the RL model:
1. After completion of each candle, mark its location in the overall structure and take trades (both Buy & Sell)
2. Close the trades at the end of next candle.
3. Define the rewards i.e positive value if its a winner and negative value if its a loss.
4. Now let the RL model iterate over the entire data and learn by itself which are the optimal candles to buy and sell so that its reward points are maximized
5. Manually this is not possible but with computer after a million odd iterations the model will get trained to identify the best candles which will give us higher points.
That's all...enjoy...
Zerodha provides a very nice python kite connect api package which can be easily used with any python backtester and it streams live data and we can place trades in real time. Also they provide historical data for backtesting.
Try Using a python open sourced packages as it will make our life easy, for this system I have used the python reinforcement learning framework, RL is a supervised machine learning algorithm which can be trained and later it will learn by itself. This algorithm is usually used in training computers to play chess, crawl web, sentiment analysis etc
You can check this article for an overview:
https://neptune.ai/blog/best-reinforcement-learning-tutorials-examples-projects-and-courses
Example trading scenario:
https://towardsdatascience.com/deep-reinforcement-learning-for-automated-stock-trading-f1dad0126a02
I will just give a hint on how I have built this system, first just look at charts and analyze how you would need to trade it manually:
Preprocessing steps:
1. Select the time frame to trade say 5 min.
2. Identify all the candle pivots (3 candles, with lower highs or higher lows on either side)
3. Mark the structure for all the identified pivots.
4. If needed you can also use some technical signals.
Analysis rules and training the RL model:
1. After completion of each candle, mark its location in the overall structure and take trades (both Buy & Sell)
2. Close the trades at the end of next candle.
3. Define the rewards i.e positive value if its a winner and negative value if its a loss.
4. Now let the RL model iterate over the entire data and learn by itself which are the optimal candles to buy and sell so that its reward points are maximized
5. Manually this is not possible but with computer after a million odd iterations the model will get trained to identify the best candles which will give us higher points.
That's all...enjoy...
You indicated that you would use 5 min time frame.
For a candle pivot, you needed 3 candles.
You are looking for a million odd iterations.
There are just 75 number of 5 minute candles in a day on the index. So, the number of pivots are far lesser than that.
Assuming 21 working days in a month, you get 252 working days in a year.
How many years of data does one need for this process to complete 1 million iterations?
Have you completed the run for such a process?
