Dev Diary 4

Welcome to dev diary 4! The diary celebrating release day, and the last diary covering the market conditions in the demo: Advanced difficulty with a moderate theme (Advanced/Balanced). The plan for future dev diaries is discussed at the end.

Deciphering the Advanced Market-Condition in the Demo

In contrast to previous diaries, the market-specific AI has a more discernible strategy in this condition because it primarily revolves around conditions on turn 1: (1) it will put most money into a stock close to $100, and (2) it buys one of the cheaper stocks, if able. On subsequent turns, the market-specific AI usually buys one of the two stocks bought on the first turn, if able.

Importantly, as illustrated in the screenshot, the latter observation is not a strict logical condition. The AI sometimes buys the cheapest stock. Other times it buys a cheaper, but not cheapest, stock.

Steam post imageThis behavior suggests that the decision depends partly on price but also depends on something unknown. Regardless, the two observations provide a consistent clue to motivate further investigation: stock price at the start of the game likely plays an important role in the underlying truth.

Looking at the relationship between stock price and future price changes looks flat on first look. However, two peculiar observations are noticeable: (1) the mean change is slightly higher at low stock prices, and (2) artificial-looking lines seem to project out of the cloud of points. These observations justify further investigation.

By default, the analysis of simulations includes data across all turns. However, because the market-specific AI seems to make game-long decisions on the first turn, the relationship on the first turn is likely the most important. Thus, using the “subset on turns” functionality in the analysis tool, we can observe that the relationship changes when only the first turn is included.

Steam post imageSpecifically, the change is highest for prices close to 100, aligning with the behavior of the market-specific AI. Because no stock has a clearly larger average change in stock price, a potential trading strategy would choose a stock close to $100 with a high dividend. This observation helps elucidate one aspect of the market-specific AI strategy, but the reason behind buying a cheap stock on the first turn remains unclear.

Since price is not explaining the reason for buying cheap stocks, the next natural predictor is change (or percent change) in price, because the other predictors do not naturally imply “buy a cheap stock” but rather focus on a different measure. Focusing again on the relationships of the first turn, change in stock price over 1 turn is not noticeably predictive of changes.

Steam post imageHowever, after a fair amount of trial and error of increasing the number of turns one-by-one and checking multiple stocks, a small “U” relationship starts to consistently appear when the number of turns increases. This relationship is particularly noticeable for changes over 4 turns and especially for the QE stock below.

Steam post imageTaken together, these observations suggest a possible trading strategy:

• On turn 1:

  1. Buy stocks close to $100.

  2. If money is left over, buy stocks that decreased more than $20 over the prior 4 turns.

  3. For both, prioritize stocks with high dividends over stocks with low dividends.

• After turn 1:

  1. Buy stocks already in the portfolio, prioritizing stocks with high dividends over stocks with low dividends.

  2. If not possible, buy bonds with the intent to sell on the next turn.

Unlike the easy and moderate difficulty, this trading strategy is the most complex trading strategy so far with 26 rules. In general, this complexity is representative of advanced market conditions because the underlying truth affects all stocks and requires contextual decisions.

Steam post imageHowever, the complexity achieves better average performance than the market-specific AI!

Steam post image Goal of the dev diaries going forward

This relatively simple game was fun to develop. The best parts were winning nail-biting games during play testing and designing the different market conditions. The dev diaries going forward will bring readers behind the curtain on the latter. Specifically, we will focus on the development of market conditions with a nightmare difficulty, and the issues faced during the development. Lastly, while a weekly diary is the target, the next several months are very busy and some weeks may be skipped.

Ideas for scenarios, game functionality, or technical issues are welcome at jerzyngames@gmail.com.