Summarizing human opinion by AI for finance applications
One of the approaches to have efficient portfolio management is to collect different expert opinions about the percentage assigned for each of the asset classes as well as asset allocation for the portfolio. In the past, portfolio managers in the asset management companies relied on their network to try to stay on top of the market and be able to achieve reasonable results for their clients based on the client risk level.
This article reviews some of the available methods to make the mentioned procedure automated through natural language processing techniques. Automated summarizing the expert opinion could be one of the steps toward democratizing the wealth management for all citizens without considering their portfolio size.
The first step for data collection involves summarizing the expert opinion and basically extract the main points from the text. The reviewed technique used the Reddit TL data set and then applied to the CNN/Dailymail news article to check the performance of the algorithm. The goal of the reviewed article is to make aligning AI systems with human preferences a central component of AI research and deployment particularly for the finance domain.
The result of the analysis demonstrates that the human feedback outperforms the supervised techniques with significant differences.
Currently, the pretrained models does not provide accurate answers in majority of the scenarios and it is highly dependent to the factors such as quality of the initial human text. When the initial quality of the trained text is low and it contains the harmful social bias the output is not as accurate as it should[2–5].
The following section demonstrates the outcomes of the human feedback, supervised and pre-trained models for the following text:
Recommended I/O grad programs that may be overlooked from searches. Optional to answer last question pertaining to MBA decision[6].
Human feedback model:
I’m in junior standing and considering going to grad school for I/O psychology. What are some good schools to look into for I/O programs? I’m in California.
Supervised model:
I’m in junior standing and I’m not sure what schools are good for I/O programs. I’m also not sure if I should go to grad school or get a job.
Pre-trained model:
I’m in junior standing right now where I’m supposed to figure out what I need to do with my life (grad school vs full-time job) and so I intended to go to grad school for a Masters and be done with school.
Finance application:
In the portfolio management world, one of the main tasks of the wealth managers and portfolio managers are to make a decision and be right. Traditional portfolio managers apart from doing the economic analysis and following the news, need to challenge their opinion based on other experts opinions. It is important to stress test the decision by checking with other opinion leaders and experts in the filed for the important portfolio allocation to maximize the possibility of achieving the customer portfolio goals. The dot analysis tool can be used to put the opinion of each portfolio manage in vote and get the rating from the pre-determined experts. Then, the weighting of each portfolio manager need to be adjusted based on their past performance to achieve the accurate consensus that will possible lead to better performance for the customers in the future decision making processes.
References:
[1] https://openai.com/blog/learning-to-summarize-with-human-feedback/
Maynez, J., Narayan, S., Bohnet, B., & McDonald, R. (2020). “On Faithfulness and Factuality in Abstractive Summarization..” arXiv preprint.
[2]Sheng, E., Chang, K. W., Natarajan, P., & Peng, N. (2019). “The woman worked as a babysitter: On biases in language generation.” arXiv preprint. [3]Bordia, S., & Bowman, S. R. (2019). “Identifying and reducing gender bias in word-level language models.” arXiv preprint.
[4]Nadeem, M., Bethke, A., & Reddy, S. (2020). “StereoSet: Measuring stereotypical bias in pretrained language models.” arXiv preprint.
