Look at beauty cosmetics from the inside, from ancient times to the present, through time and space.

Beauty products have both efficacy and spiritual consumption attributes, and the triple values of time, research and development and scale support the valuation premium of international beauty leading groups. The history of China’s beauty industry is still short, and it is in the stage of gaining momentum from marketing/channel-driven to R&D-driven, improving comprehensive strength and transforming into a real consumer brand, with a broad market and rich opportunities; The multi-brand and collectivization trend of local high-quality beauty cosmetics leaders has begun to appear, and the growth supports high valuation.

▍ The valuation premium of the beauty industry is due to the value of spirit and efficacy, and the triple value of time, technology and scale breeds the valuation premium.

The historical P/E center range of the international beauty industry leader is 20x-35x, which generally enjoys a valuation premium compared with its index. The P/E multiple generally exceeds 1.4, and the P/S multiple generally exceeds 1.3. We believe that the valuation premium enjoyed by international beauty leaders is due to:

1) The value of time: the precipitation of time creates brand value. International beauty leading companies have a long history of more than 70 years, and their brands are well-known, rich in connotation and enjoy a premium.

2) The value of science and technology: The international beauty leading group has invested heavily in basic research (skin texture/raw materials, etc.), applied research, safety testing and other aspects for a long time to build tangible technical barriers and intangible consumer trust barriers.

3) Value of scale: collectivization exerts brand synergy, globalization captures dividends in various markets, leading the scale and stabilizing the position of the industry.

▍ China’s beauty industry enjoys a valuation premium, with a broad market and abundant opportunities.

The valuation premium of China’s beauty industry has existed for a long time, with the P/E multiple center of 2.3 and the P/S multiple center of 3.1. Main reasons:

1) Early stage of development: low permeability and leading growth rate. In terms of penetration rate, the per capita consumption of beauty care in China is only about 403 yuan, far lower than that in Europe, America, Japan and South Korea. In terms of growth rate, CAGR+10.1%% in China’s beauty care market from 2008 to 2021 is higher than CAGR+2.6%% in the world.

2) Continuous iteration, full of vitality, driven by channel to technology/R&D.. China’s beauty industry has a fast iteration: consumers are becoming more rational, supervision is becoming stricter, channels are more dispersed, traffic costs are constantly high, and marketing focuses on value recognition.

▍ China’s beauty industry has a short history and is full of vitality.

The development of local beauty listed companies started relatively late, and most of their brands were launched after 2000. The market concentration of beauty personal care in China is still low. In 2021, the proportion of CR3/5 is 26.8%/33.5% and there are no local companies in CR5, which is lower than that in Europe, America, Japan and South Korea. There is still room for improvement in the market share of high-quality local leading companies. At present, the industry is poised for stage adjustment, from channel and marketing-driven to technology-driven, and companies with strong comprehensive capabilities are expected to further increase their market share.

▍ The pattern is scattered, high-quality leading players enjoy industry growth and share increase, and leading players in sub-sectors rise.

▍ Risk factors:

Overseas historical development is not fully applicable to China beauty listed companies; The weak consumption of optional consumption, especially in the field of beauty, affects the company’s performance: the industry updates quickly, and the company, brand or product may have a certain life cycle; Industry competition or intensification, affecting the company’s performance; The sales volume of new products may not meet expectations, and the construction of brand matrix still has a long way to go.

▍ Investment strategy:

After the valuation adjustment since the second half of 2021, the PE of the leading A-share beauty companies in 2023 is concentrated in 45x~55x, which has fully reflected the factors such as the slowdown of industry growth, the ready research and development, and the shift of driving force. In short, in 2023, with the gradual recovery of consumption, the efficiency improvement of various companies’ operations/the promotion of large single product strategy/the listing of new products, there is still room for improvement in the valuation of leading companies; From 2023 to 2025, with the improvement of R&D and comprehensive strength of local brands, as well as the improvement of consumers’ recognition of local brands, the beauty industry is expected to change from marketing-driven and channel-driven to R&D-driven, and high-quality bibcock is expected to become a real consumer brand with both product power/marketing power/channel power, thus maintaining a high valuation.

This article comes from a selection of brokerage research reports.

Infrastructure for training AI to solve common problems

In order to train artificial intelligence models that can solve common problems, infrastructure is needed to provide support. These infrastructures are usually composed of hardware, software and tools to improve the efficiency and accuracy of model training. This article will introduce the infrastructure for training AI to solve common problems.

I. Hardware infrastructure

When training artificial intelligence models, it is usually necessary to use high-performance computing hardware to provide support. The following are several common hardware infrastructures:

  1. CPU: The central processing unit (CPU) is a general-purpose computing hardware, which can be used to run various types of software, including artificial intelligence models. Although the performance of CPU is relatively low, it is still useful in training small models or debugging.

  2. GPU: A graphics processor is a special computing hardware, which is usually used to process images and videos. Because of its highly parallel structure, GPU can provide higher computing performance than CPU when training artificial intelligence models, so it is widely used.

  3. TPU: Tensor processor is a kind of hardware specially used for artificial intelligence computing, developed by Google. The performance of TPU is higher than that of GPU, and it is suitable for large-scale artificial intelligence model training and reasoning.

Second, the software infrastructure

In addition to hardware infrastructure, some software tools are needed to support the training of artificial intelligence model. The following are some common software infrastructures:

  1. Operating system: Artificial intelligence models usually need to run on an operating system, such as Linux, Windows or macOS.

  2. Development environment: Development environment usually includes programming language, editor and integrated development environment (IDE) for writing and testing artificial intelligence models. Common development environments include Python, TensorFlow, PyTorch and Jupyter Notebook.

  3. Frames and libraries: Frames and libraries provide some common artificial intelligence model algorithms and data processing tools, making model development and training more convenient. Common frameworks and libraries include TensorFlow, PyTorch, Keras and Scikit-Learn.

Third, the tool infrastructure

In addition to the hardware and software infrastructure, some tools are needed to support the training of artificial intelligence models. The following are several common tool infrastructures:

Dataset tool: Dataset tool is used to process and prepare training datasets, such as data cleaning, preprocessing, format conversion, etc. Common data set tools include Pandas, NumPy and SciPy.

2 Visualization tools: Visualization tools are used to visualize the training process and results to help users better understand the performance and behavior of the model. Common visualization tools include Matplotlib, Seaborn and Plotly.

Automatic parameter tuning tool: The automatic parameter tuning tool is used to optimize the parameters of the model to improve the performance and accuracy of the model. Common automatic parameter tuning tools include Optuna, Hyperopt and GridSearchCV.

In short, training artificial intelligence models to solve common problems requires the use of a variety of infrastructures, including hardware, software and tools. These infrastructures are designed to improve the efficiency and accuracy of model training, so that the model can better solve various practical problems. In practical application, users need to choose the appropriate infrastructure according to specific requirements and data characteristics, and design and implement it accordingly.