the music in state capitalism: working with the regulations in China

In his book Money Machine, Shan Weijian describes the ups and downs of working within China’s regulatory system in the 2000s. I found it insightful to observe several recurring behavioral traits.

Unless otherwise specified, “the bank” refers to Shenzhen Development Bank, which Newbridge acquired a controlling stake in 2004 and sold to PAIG in 2011.

1. Principles of social engineering

I had the impression that officials adhered to strong, often top-down, social engineering principles and used them to guide even the finer details of transactions.

For example, when Baosteel (a highly profitable state-owned steel company) expressed interest in acquiring the bank, regulators shut it down because, in principle, an industrial raw-materials company was not allowed to own a bank. While such structures exist elsewhere (for example, Mitsubishi owns MUFG in Japan), current regulations in the US and EU prohibit this to prevent banks from becoming “internal treasuries” for industrial firms.

2. Intense leverage and even threats

When the government pursued the share reform in 2005, what began as voluntary participation quickly became coercive. Out of concern that the release of legal-person shares would flood the market and depress prices, a common arrangement required participating companies to grant additional “free shares” (about 30% of the traded amounts) to public shareholders.

Newbridge chose not to participate, worrying that this arrangement would significantly dilute its roughly 20% stake. In response, stock exchanges—under pressure to advance the reform—effectively blocked the bank from releasing its quarterly financial statements, forcing it to violate disclosure requirements for listed companies.

Note: China’s 2005 share reform aimed to make historically non-tradable legal-person shares tradable on stock exchanges.

3. Multiple, often competing, regulations

To obtain approval for the deal with PAIG, both parties had to navigate multiple layers of government—from municipal to provincial authorities, and ultimately Beijing—while also dealing with sector regulators such as the China Securities Regulatory Commission and the China Banking Regulatory Commission.

At best, these regulatory requirements were overlapping; at worst, they were contradictory. For instance, when the bank sought to issue new shares to raise capital, banking regulators supported the move because it strengthened the bank’s capital base and reduced risk. Securities regulators, however, opposed it, arguing that an increased supply of shares would depress the stock price.

4. Publicity, “face,” or pride.

Anticipated public perception is a critical factor in decision-making, because unfavorable perceptions could be detrimental to one’s “face,” or career. When Newbridge filed an arbitration in Paris against the Shenzhen government for breaching the binding transaction agreement, it cracked the government’s once-firm stance on not proceeding with the case. There were signals suggesting that Shenzhen was willing to resume talks in exchange for a withdrawal of the arbitration.

I’d like to note that this does not mean a simple “talk-or-sue” strategy works when dealing with Chinese regulators. In fact, the most important factor may have been that certain top-level officials were still interested in seeing the case go through. Leveraging that momentum, together with the pressure created by the publicity surrounding the arbitration, contributed to the thawing of the negotiation impasse.



First draft: Jan 11, 2025

Second draft: Jan 14, 2025

a Korean lesson: arts of negotiation

The Art of Negotiation
Recently, I’ve been reading Shan Weijian’s memoir about leading Newbridge Capital’s acquisition of Korea First Bank. I was deeply inspired by the techniques and artistry he demonstrated in negotiation. Below are a few principles I’ve distilled to help form a more systematic understanding.

1. Always Think in Terms of Bottom Lines
All of Newbridge’s proposals were built on one fundamental principle: the investment had to make commercial sense. Based on this, clear red lines were established—rules that could not be crossed or compromised. For example, when the government demanded that Newbridge’s investment return not exceed 12%, Newbridge flatly refused.

2. Understand the Consequences of Concessions Through Clear Calculations
Should non-performing loans be priced at 95% of face value or 96%? What does that 1 percentage point concession mean for overall returns? Only with a clear, quantified understanding can you know how many cards you hold and how much room you truly have to maneuver. An operational, flexible financial model is a must.

3. Play Your Cards One at a Time
If you can afford to concede three dollars, start by conceding one. Save the remaining two as bargaining chips for when you need leverage later in the negotiation.

4. The Role of Agents as “Messengers”
Agents and intermediaries serve two key purposes:
They help establish trust between previously unfamiliar parties and bridge differences in position (for example, diplomats or ambassadors acting as intermediaries between foreign investors and financial regulators).
They create an informal “buffer layer” through which tough messages can be delivered without triggering immediate or irreversible consequences.
For instance, Newbridge’s “ultimatum” in September 1999 could not be formally delivered to the government—politically, a government yielding to a foreign investor’s ultimatum was unacceptable. Instead, the message was conveyed via an ambassador: “This is the last opportunity. If it’s not signed now, there will be no chance to sign later.”

5. Managing Public Opinion and Lobbying Politicians
Winning the media battle can be important, but it often creates only localized pressure, and the direction of public opinion is hard to control. Shan maintained regular contact with journalists, which not only helped shape public perception but also provided insights into political dynamics and the progress of a parallel bank acquisition. At the same time, this engagement helped establish Newbridge’s image as professional, rational, and commercially driven in the eyes of the media and the public.

6. Understand Your Counterparty’s True Interests
For financial regulators, the political and regulatory implications of certain agreements may matter more than the precise monetary amounts involved. Once Newbridge recognized this, it could adjust the form of the terms—without crossing its own bottom lines—to better satisfy the counterparty’s political needs, while leaving its financial returns intact.

7. Be Patient: Sugar-Coat Your Punches and Threats

Expressing firmness without breaking the negotiation is an art, especially when the other side makes unreasonable demands.

Instead of saying, “You’ve gone back on your word and broken your promises!”, say:

“We are confused and disappointed by your new proposal, which represents a significant step backward from the Hong Kong discussions that took place just one day ago.”

Instead of saying, “We’ve invested so much time and resources in this negotiation—we won’t walk away empty-handed, and we’ll sue if we have to,” say:

“We entered this process in good faith based on the government’s commitments, and therefore invested substantial time and resources to complete the transaction. Our investors will not allow us to walk away quietly. We will do our utmost to protect our rights, though we hope to avoid a mutually destructive outcome.”

Negotiation, at its core, is not just about toughness or compromise—it is about clarity, patience, and the disciplined use of leverage.

why supervised fine-tuning is infeasible

We hoped that fine-tuning is best-suited for a niche purpose. However I don’t see it as the future of adapting generative AI in industry. Here’s why:

1. Dependency on high-quality annotation. Cliche, but it is hard and expensive to get. The process is not always controllable, and often very time consuming (e.g. the annotation team may mis-interpret the annotation guideline)

2. The foundational capability of LLMs are only getting better. With appropriate prompts, it could solve a large chunk of problems.

3. Fine-tuning can make a model worse. If the training data contains misleading / bad signals, it could actually mislead the models and “erase” some alignment / intelligence originally possessed by the foundational model. A version of “garbage in, garbage out”.

4. Human cost is much higher compared to the computational cost. With tools and techniques (e.g. LoRA, llama-factory), the computational cost of fine-tuning is actually low; however, the human time involved in updating the annotation guideline, debating the wordings and standards, etc are the larger cost.

5. The marginal cost of obtaining new annotations is just much higher than updating the prompts. Fixes an imperfection in annotation guideline could take weeks to turn around. Fixes on prompts, by contrast, could be much lower.

6. It is much easier to manage a library of prompts compared to a library of fine-tuned models.