Skip to content

How to write prompts

When using LLMs to generate trading strategies, it is important to write clear and concise prompts that accurately describe the desired strategy. Here are some tips and examples to help you write effective prompts.

Mandatory fields

In every strategy, there are 4 mandatory fields that need to be specified, and those are:

  • ticker: The ticker symbol of the stock or asset you want to trade. For example, "AAPL" for Apple Inc. You can find those for every equity on Yahoo Finance.
  • position_type: The type of position you want to take. This can be either LONG, SHORT, or LONG_SHORT_COMBINED. If you do not specify this, the default value is LONG.
  • buy_condition: The condition that triggers a buy signal. This can be any of the conditions available in the library and can be combined using logical operators.
  • sell_condition: The condition that triggers a sell signal. This can be any of the conditions available in the library and can be combined using logical operators.

Those fields are mandatory, and the best practice is to start with them. So the first sentence of a prompt should be something like:

Can you please generate a strategy for AAPL where you buy when RSI moves 
above 30 and you sell when RSI falls below 70.

Simple and clear. The model will understand that you want to create a strategy for AAPL and that you want to buy when RSI moves above 30 and sell when RSI falls below 70. Position type is not specified, so the model will use the default value which is LONG.

Another and more complicated example of a prompt with mandatory fields:

Could you outline a long strategy for NVDA that sells off when the Ichinoku 
Conversion oscillates by 19.9 percent over 43 days or ATR changes by 95.02 
percent over 45 days and take a long position when the TRIX where length 
is equal to 26 varies by 13.6 percent within a interval.

Let's break this down:

  • ticker: The ticker symbol is defined right at the beginning of the prompt, so the model will know that you want to create a strategy for ticker NVDA.
  • position_type: The position type is defined as long in the prompt, so the model will know that you want to create a LONG strategy.
  • buy_condition: The buy or long condition is defined as a second condition this time and it is:
take a long position when the TRIX where length is equal to 26 varies by 
13.6 percent within a interval.

This is still just one condition that can be exactly defined in the library as:

IntraIntervalChangeOfXPercentCondition(series=TRIX('NVDA', length=26), percent=13.6)

If you want to specify parameters of an indicator as in this example of TRIX, you should do that right after the name of the indicator. So in this example the model will know that you want to create a TRIX indicator with length equal to 26.

  • sell_condition: The sell condition is defined as a first condition this time and it is:
sells off when the Ichinoku Conversion oscillates by 19.9 percent over 
43 days or ATR changes by 95.02 percent over 45 days

This is a combination of two conditions that are separated by or operator. The final condition will be:

OR(
    ChangeOfXPercentPerYDaysCondition(
        series=ICHIMOKU_CONVERSION('NVDA'),
        percent=19.9, 
        number_of_days=43
    ), 
    ChangeOfXPercentPerYDaysCondition(
        series=ATR('NVDA'), 
        percent=95.02, 
        number_of_days=45
    )
)

As you can see all the parameters are specified in the prompt and the model will know that you want to create a ChangeOfXPercentPerYDaysCondition condition with the parameters specified in the prompt. If you don't specify parameters for indicators, the model will use default values for them.

Optional fields

In addition to the mandatory fields, there are several optional fields that you can include in your prompt to customize your strategy further. You include them after the main sentence about conditions of entry and exit. Here are some examples of optional fields:

stop_loss and take_profit

You can specify the stop loss and take profit for the strategy. For example:

Apply stop-loss at 42.56 percent. Apply take-profit at 25.16 percent.

start_date and end_date

You can specify the start and end date for the strategy. For example:

Set the start date to 2004-10-28. Set end of the strategy to 2024-10-4.

It's recommended to specify the date format as YYYY-MM-DD so the model will understand it better because this was the format of the training data.

interval

You can specify the interval for the strategy. For example:

Set the interval equal to 1 week.

period

You can specify the period for the strategy. For example:

Set the period to 6 months for the strategy.

If the period is specified, the model will ignore the start and end date.

initial_capital

You can specify the initial capital for the strategy. For example:

Set the initial capital to 1000 dollars.

order_size

You can specify the order size for the strategy. See what order sizes are available at OrderSize documentation. For example:

Set order size per trade to 20 dollars.

This will set: order_size=USD(value=20).

or

Set order size to 2 contracts.

This will set: order_size=Contracts(value=2).

trade_commissions

You can specify the trade commissions for the strategy. See what trade commissions are available at TradeCommissions documentation. For example:

Set trade commissions to 0.01 percent.

This will set: trade_commissions=PercentageCommissions(value=0.01).

or

Set trade commissions to 0.5 dollars.

This will set: trade_commissions=MoneyCommissions(value=0.5).

This is how you can write prompts for LLMs to generate trading strategies. Quick recap:

  • Start with mandatory fields: ticker, position_type, buy_condition, and sell_condition. The best practice is to set them in the first sentence.
  • Define optional fields after the first sentence. If you do not specify them, the model will use default values for them that you can find here.

Examples of full prompts

Can you please generate a strategy for AAPL where you buy when RSI moves 
above 30 and you sell when RSI falls below 70. Apply stop-loss at 10 percent. 
Set take-profit at 5 percent. Set the initial capital to 10000 dollars.

or 

Could you outline a long strategy for NVDA that sells off when the Ichinoku 
Conversion oscillates by 19.9 percent over 43 days or ATR changes by 95.02 
percent over 45 days and take a long position when the TRIX where length 
is equal to 26 varies by 13.6 percent within a interval. Set the start date 
to 2022-1-1 and set end of the strategy to 2024-1-1. Set trade commissions 
to 0.01 percent.

Important note at the end

It is highly recommended to work with the documentation of a Strategy object to know what parameters are available for each condition and how to use them. You can also check the TradingSeries documentation to see what parameters are available for each indicator. The model will understand the prompt better if you specify the parameters in the prompt.

If you are not sure how to write something in the prompt, you can look at the file at this link. There are some options on how to specify the parameters in the prompt. If you use the same patterns as in the file, the model will understand it better because those are the patterns that were used for training the model.