Bridgewater Associates' believability-weighted system for algorithmic decision-making

There are moments you never forget: Like Ray Dalio with his hard New Yorker accent describing the meritocratic ranking system at Bridgewater Associates where the input of his analysts gets weighed and that weighing factor changing over time.

I like to believe that the hard New York English comes close to the accent when Germans speak English, others say we sound more like Amish with their Pennsylvania Dutch.

Ray Dalio and Bridgewater Associates

Bridgewater Associates is an American investment management firm founded by Ray Dalio and known for its distinctive corporate culture where radical transparency and algorithmic decision-making are the norms. Bridgewater Associates’ decision-making system is deeply engrained in their corporate culture.

Ray Dalio has described their system as “believability-weighted decision making”. The idea is that every person's opinion in the firm is considered, but not all opinions are considered equally. Instead, they are weighted by the person's credibility on the topic at hand. This initial expertise becomes credibility over time, or “believability”.

Believability is determined algorithmically based on the person's track record of making good decisions in the relevant area. The assessment of accuracy of individuals' “believability” or expertise can also change over time when the algorithm is changed to increase fairness and bias.

The merit of the input is all that counts

This meritocratic system aims to prevent decision-making from being influenced by hierarchical position or personal bias, and instead be based on the merit of the ideas themselves and the proven expertise of the person suggesting them.

Critique and Limitations

Critics argue that it could lead to a lack of diversity in thought, as those with higher believability scores have a disproportionately large influence on decisions. There are also other critiques.

But most are solved by following the standard mitigation practices like,

• Ensuring a diverse team to provide a broader range of perspectives.
• Implementing anonymous input.
• Regular review and adjustment of the algorithm that calculates believability scores.

Any system of decision-making, including a believability-weighted system, will inherently have some limitations. The key is to recognize these limitations and to continually strive to mitigate them as much as possible.

One thing such mitigation strategy could be to balance the believability-weighted system with other decision-making approaches.

Balancing with Other Decision-Making Approaches

The believability-weighted system could be balanced with other decision-making approaches. But balancing different decision-making approaches can easily lead to more complexity and potentially costly in terms of time and resources.

However, there might be situations where this approach could be beneficial. For instance, in certain high-stakes decisions, like managing million dollar investments. Or in situations where the team is deeply divided, it might be beneficial to use a different decision-making method.

That said, the decision to employ such a strategy would depend on the specific context and the potential benefits and drawbacks.

Conclusion: Parallels in Assigning a Measure of Certainty or Weight to Information

Bridgewater Associates' believability-weighted system for algorithmic decision-making reminded me of the confidence that you ascertain when you gather and collect data from individuals to increase data validity.

My research on the specifics of Bridgewater Associates algorithm and the factors it takes into consideration however aren't publicly disclosed. Bridgewater maintains that their system is proprietary and part of their competitive advantage.

So all I can state is that both concepts—Bridgewater's believability-weighted decision-making and the concepts of self-reported confidence—deal with assigning a measure of certainty or weight to information based on past data or performance.

In Bridgewater's system, the “believability” score given to an individual's opinion is based on their track record in the relevant area.

Similarly, the concept of confidence we can gain additional insights that might not be captured by the data alone. By asking individuals about their confidence level when providing a data point or ranking, should provide us with indications of how certain we can be that a given range contains the true data point when we compare their confidence in a panel study over time.

So, in both cases, the aim is to make better-informed decisions by using past data to determine the weight or confidence we give to a particular piece of information or opinion.

In my ongoing research project, I want to ask individuals how confident they are in providing a certain qualitative datapoint or quantitative ranking.

If you want to stay up to date, consider subscribing to my newsletter.