Dev Diary 9

Okay! This week will focus on the development of a market-specific AI for the 2nd nightmare scenario. As discussed last week, all stocks follow a sin function; that is, a wave-like function that goes up and down. The timing of the increase depends on two features during the run-in period:

• The price difference between the beginning and the end of the run-in period

• The difference between the highest and lowest prices during the run-in period

As discussed in Dev Diary 7 (link), a market-specific AI aims to provide insight into the underlying truth and beat the default AIs, it is not intended to implement the most optimal trading strategy. In this context, our first attempt at the market-specific AI will implement the following high-level trading strategy:

1. On turn 1: Invest in a stock with the following characteristics from the run-in period: (1) did not change much with (2) stable prices.

   • If the two features sum-to-zero, then a stock with these characteristics should be on the step part of the sin curve. This approach to stock selection is necessary because these are the only characteristics predictive of returns available on turn 1 for selecting stocks.

2. After turn 1: Invest in stock(s) with notable increases in price from the prior turn.

   • While working through the implications of the run-in period characteristics would more accurately predict the stocks providing higher overall expected returns, a focus on buying the stocks starting to increase after declining for several turns may provide a simple but effective approximation.

On a practical level, the trading strategy on turn 1 can be implemented by requiring a stock to go up by, for example, -20 and go down by -20. However, as illustrated in the screenshot below, the game currently does not allow negative numbers. Steam post image

Although the issue was relatively straightforward to fix, which will be pushed when the nightmare scenarios are released.

We only implemented the turn 1 strategy for stocks with relatively low volatility, which minimizes the impact of the minimum and maximum components of the shift. Additionally, as illustrated below, we bought only a percentage of each stock given the uncertainty in not formally integrating the shift into the strategy.

After the 1st turn, we buy stocks that increased the prior turn but decreased over the prior 4 turns. The latter requirement attempts to ensure that the stock was recently near the bottom portion of the sin curve. As a small optimization, we did not include the THC stock at all due to its high natural volatility (that is, the higher volatility may result in a higher probability of randomly meeting these conditions). Surprisingly, this relatively simple trading strategy performed well and will be the market-specific AI for this market condition.

Going forward, I will create the last Nightmare market condition off-line, which will ensure players don’t know the underlying truth for each market condition with certainty. The game will likely be updated in July after a family vacation. I hope people enjoyed the dev diaries and the game as much as I have enjoyed making them.