Can machines completely replace humans in stock investment?

Since the beginning of this year, in the context of the massive outbreak of artificial intelligence technology, quantitative and AI strategies have performed well and have attracted market attention with their advanced algorithms.

By the end of 2022, as small-cap stocks gained momentum, traditional active equity funds showed signs of delisting, and fundamental investment trends faltered. At the same time, A-shares have experienced an overall slight decline since the beginning of this year, with various broad-based indices experiencing varying degrees of decline, and equity fund indices performing even weaker.

Compared to active equities, quantitative and index-enhanced products have outperformed in the market under the year's volatile conditions. Among them, a group of quantitative strategies that rely entirely on artificial intelligence for stock selection have attracted market attention with their investment performance. Does this imply that we are entering an era where machines will completely replace humans in the field of investment?

Shenwan Hongyuan believes that, at present, it is recommended that active equity investors "stick to fundamental investment and wait for the return of fundamental market conditions." Historically, the effectiveness of volume-price indicators during fundamental market conditions will quickly decrease, and even the effectiveness of technical indicators will decline. There is a cyclical relationship between the two, and currently effective volume-price indicators may weaken cyclically in the future, which is when fundamental investment is most suitable.

What are the advantages of AI investment?

In a report on September 4th, the Shenwan Hongyuan Denghu team mentioned the following characteristics of AI stock selection strategies in terms of technical principles:

(1) Opacities. In AI strategies, the work of fund managers is more focused on algorithm frameworks, parameter tuning, preventing overfitting, turnover and exposure limits, and tracking management. The specific logic of stock selection is not reflected by the fund managers.

(2) High threshold. In fact, the threshold for using AI strategies is not high at the current stage. However, the low threshold for widespread use of AI strategies will, in turn, raise the threshold for AI strategies. Currently, using mature algorithms for AI strategy development does not require investment strategies to fully understand the mathematical models of the algorithms. Just being able to use them can develop strategies. In the current market environment, the effectiveness of AI strategies will inevitably stimulate more teams to try to develop AI strategies. Ultimately, relatively simple frameworks may become ineffective, and further refinement of details and mastery of more advanced algorithms may become essential weapons for investment teams, ultimately raising the threshold for using AI strategies as a whole.

(3) Profiting from volume and price. AI stock selection strategies mainly profit from volume and price. This partly explains the strong performance of AI strategies in recent years. The outstanding performance of the strategies has its own advantages and is closely related to the market environment with effective volume and price.

However, this does not mean that AI stock selection strategies are superior to humans. According to the Denghu team's model calculations and performance since the beginning of this year, AI strategies have shown the following important characteristics within historical intervals:

1. When it comes to investment, AI behaves more like a human, sometimes making significant adjustments to its investment style, even attempting to predict turning points in advance, and sometimes being able to maintain its style in trending markets without obvious regular features.

2. The flexibility of AI strategies is a double-edged sword: From the perspective of market value style alone, the returns of AI strategies come from both the trend of market value style and the attempt to predict inflection points and style reversals. If an AI strategy experiences a drawdown, it may be difficult to understand the drawdown.

3. AI strategies have unique advantages in handling volume and price, with variable sign direction, significant adjustments, and flexible tracking of market trends and reversals, which are difficult for traditional linear volume and price factors: AI strategies have unique advantages compared to traditional multifactor volume and price usage. Taking market value style as an example, the use of traditional market value factors often only exposes small market values in a single direction, and often encounters market fluctuations, leading many quantitative teams to pursue market value style neutralization. However, AI strategies have variable sign directions in market value style, with smaller market values when small market values are performing well, and larger market values when large market values are strong. They also frequently make significant adjustments and attempt to predict inflection points, which is difficult to achieve in the traditional multifactor framework of momentum bias factors.

AI Strategies vs. Fund Managers: Each Has Its Own Strengths

Currently, the Denghu team believes that the respective advantages of AI strategies and active equity fundamental investments are similar to the advantages of quantitative and active fundamental investments. AI strategies are better at dealing with volume and price, as well as small stocks closely related to volume and price, while active fundamental investments excel at focusing on fundamentals and large stocks closely related to fundamentals.

If AI strategies are regarded as a large-scale composite volume and price factor, the impact of their black box nature can be reduced to some extent in their usage. In the usual multifactor framework, the selection and weighting of factors are often based on the performance of factors in the previous period, essentially momentum-based factor analysis. When an AI strategy becomes a factor, regardless of its own interpretability, the combination only selects or does not select the strategy based on the performance of the factor, and how much of the AI strategy to allocate. This may be a suitable way for traditional multifactor frameworks to use AI strategies.

There Is No Permanent Growth, Only Eternal Cycles

Generative AI algorithms represented by ChatGPT represent the future direction of artificial intelligence: high computing power + strong algorithms.

Therefore, the reduction of the threshold for AI strategies also depends on the improvement of computing power. In the current market environment, the most direct impact of computing power on AI strategies is that the improvement of computing power can accelerate the update frequency of AI stock selection models, thereby gaining an information advantage with updated information and transforming computing power advantages into profit advantages.

In the face of the development of AI stock selection strategies, the Denghu team analyzed in the report:

Although AI strategies may make significant progress in text recognition and fundamental analysis in the future, at present, the prominent advantage of AI strategies lies in the market environment of small and medium market capitalization and emphasis on volume and price, such as in the year 2023. Before AI strategies further evolve, this market style remains cyclical rather than sustained, making AI more suitable for cyclical environments.

One possible strategy to consider is to combine fundamental holdings with quantitative or AI strategies to provide short-term trading views based on volume and price judgments. In other words, while maintaining the fundamental views on listed companies, fundamental investments can be enhanced through some trades. The advantage of this approach is that it does not change the basic framework of active equity ownership. However, it comes with the drawback of increased trading costs, and these trades still lack a guaranteed probability of success, relying only on a large number of repetitions to provide excess returns. This may be difficult to sustain for active fundamental investors.