What happened if GPT-4 performed the risk assessment of the failed Silicon Valley Bank?



On March 10, 2023,

Silicon Valley Bank, which was based in Silicon Valley where many technology companies are concentrated and actively invested in technology companies, went bankrupt . In connection with this matter, software engineer Matt Eskridge conducted an experiment to `` let the large-scale natural language model GPT-4 released by OpenAI in March evaluate the risk of Silicon Valley banks '', and the results I am reporting it on my blog.

GPT4 and Silicon Valley Bank – Matt Eskridge
https://blog.matteskridge.com/business/gpt4-and-silicon-valley-bank/2023/03/19/

Silicon Valley Bank is a bank based in Santa Clara, California, USA, and has a strong presence in Silicon Valley, actively investing in many technology companies and startups.

However, in the financial results for the first quarter of 2023 on March 8, 2023, the bank's CEO Greg Becker announced that he would ``sell substantially all securities'' and ``$2.25 billion (about 300 billion yen). ) to raise funds. As a result, distrust spread among investors, and the stock price plummeted by more than 60%.

What kind of bank is 'Silicon Valley Bank' and why did it fail? -GIGAZINE



Regarding the failure of Silicon Valley Bank, which affected the economy around the world , there were voices of criticism from within, such as ``the CEO is too stupid'' and ``in-house technology was stagnant despite investing in cutting-edge technology.'' going up It is also reported that some venture capitalists and experts were aware of the crisis at Silicon Valley Bank several months before the bankruptcy.

The failure of Silicon Valley Bank was already foreseen as of 2022-GIGAZINE



Therefore, Mr. Eskridge gave 2021 data on Silicon Valley Bank to the language model `` GPT-4 '' released by OpenAI in March and conducted a test to evaluate the risk. GPT-4, developed as a ``multimodal AI'' that accepts not only text but also image input, has performance enough to enter the top 10% of the bar exam, and breaks through the verification system to prevent bots. , has created a new programming language and is attracting attention.

First, Eskridge gave GPT-4 the following prompt:

Consider a hypothetical scenario: the year is 2021 and the Federal Funds Rate (FF rate) is targeting 0-0.25%, similar to the past decade. A bank reported the following asset allocation in its balance sheet for the last year:

・U.S. bonds (average maturity of 10 years): 7.5%
・Mortgage-backed securities: 47.6%
・High-risk loans to companies in the high-tech sector: 31.1%
・Cash: 3.7%
・Other assets: 10.1%

Liabilities are as follows:

・Deposit currency (demand deposit): 92.5%
・Liabilities: 1.6%
・Other liabilities: 5.9%

Please rate the following risks to the bank's financial health and provide your Impact, Likelihood and RAC (Risk Assessment Code) scores for each. Please rate the degree of impact and possibility on a scale of 5 out of 5.

1: Price increase measured by CPI (Consumer Price Index)
2: Interest rate risk
3: Economic recession
4: Price fluctuations in the housing market
5: Pandemics, wars and disasters
6: Bank runs



The answer given by GPT-4 to the above prompt is as follows.

In order to assess the risks associated with a bank's financial health, we first briefly describe each risk factor and assess its impact and likelihood on a 5-point scale. Finally, for each risk, we would like to calculate a RAC (Risk Rating Code) score, which is the impact multiplied by the likelihood, to identify the risks that pose the greatest threat to the financial health of the bank.

1. Price increases as measured by CPI: Inflation reduces the purchasing power of money and can affect a bank's profitability and asset values.
Impact: 2 (low to medium), Likelihood: 3 (medium)
RAS score: 2 x 3 = 6

2. Interest rate risk: Interest rate risk refers to the possibility of incurring losses due to fluctuations in interest rates. Banks' assets and liabilities can be affected by changes in interest rates, particularly given their heavy allocation to mortgage-backed securities and long-term government bonds.
Impact: 4 (High), Likelihood: 4 (High)
RAC score: 4 x 4 = 16

3. Recession: A recession could lead to higher credit losses, especially in high-risk sectors, and lower mortgage-backed securities.
Impact: 4 (high), Likelihood: 3 (medium)
RAC Score: 4 x 3 = 12

4. Housing Market Price Fluctuations: With 47.6% of the bank's assets being mortgage-backed securities, housing price fluctuations can have a significant impact on the value of these securities.
Impact: 4 (high), Likelihood: 3 (medium)
RAC Score: 4 x 3 = 12

5. Pandemics, wars and disasters: These events can have wide-ranging impacts on economies and bank assets, but are generally considered less likely than other risks.
Impact: 4 (High), Likelihood: 2 (Low)
RAC score: 4 x 2 = 8

6. Bank runs: With 92.5% of bank liabilities in deposit currency, a bank run could have a significant impact on solvency.
Impact: 5 (very high), Likelihood: 1 (very low)
RAC score: 5 x 1 = 5

Based on the calculated RAC score, interest rate risk (RAC score: 16) poses the greatest threat to banks' solvency, followed closely by recession and housing market price volatility (both RAC score: 12). Related to



Eskridge then asked, 'What would you recommend banks do to mitigate this biggest risk?' The answer of GPT-4 is as follows.

To mitigate the most important risks identified in the previous analysis (interest rate risk, economic recession, housing market price volatility), banks can consider the following strategies.

1. Interest rate risk
a: Portfolio rebalancing: Banks can adjust their asset allocation to reduce their concentration in long-term government bonds and mortgage-backed securities. By distributing to shorter term bonds or other less interest rate sensitive asset classes, banks can reduce the impact of interest rate fluctuations.
b: Interest rate hedging……You can hedge potential interest rate fluctuations by using financial products such as interest rate swaps, futures, and options. These financial instruments help banks manage interest rate risk and limit potential losses.
c: Asset Liability Management (ALM): Introduce ALM strategies to more closely match cash flows with the sustainability of bank assets and liabilities. This ensures that interest rate fluctuations affect both sides of the balance sheet in a balanced manner, reducing interest rate risk.

2. Recession
a: Reducing the share of high-risk loans: Banks can reduce the concentration of high-risk loans in high-tech companies that are more likely to default during a recession. By diversifying their loan portfolios across different sectors and companies with high credit quality, banks can mitigate the impact of economic downturns on their loan portfolios.
b: Strengthen contract assessment standards: Tighten credit and underwriting standards for new loans so that borrowers are more likely to repay even when the economy deteriorates.
c: Maintain capital reserves: Banks need to have sufficient capital buffers to absorb potential losses during a recession. This can be achieved through careful capital planning and compliance with regulatory capital requirements.

3. Price Fluctuations in the Housing Market
a: diversify asset allocation – reduce the concentration of mortgage-backed securities in bank portfolios and distribute to other asset classes; This helps mitigate the impact of housing market volatility on the bank's financial position.
b: Monitor housing market developments: Regularly analyze housing market developments and economic indicators to identify potential risks and adjust the Bank's asset ratios accordingly.
c: Mortgage-backed securities risk management: Implement robust risk management techniques, including stress testing, scenario analysis, and credit quality monitoring, for banks' mortgage-backed securities portfolios. This will enable banks to better understand and manage the risks associated with price volatility in the housing market.

“By implementing these strategies, banks can reduce their exposure to the most significant risks and improve their financial stability and solvency,” said GPT-4.



Finally, Mr. Eskridge asked, 'It's a true or false question. Have you ever been under the control of the FDIC (Federal Deposit Insurance Corporation) because the Silicon Valley Bank failed?' GPT-4 then replied, 'False. To my knowledge as of September 2021, Silicon Valley Banks have never been seized by the FDIC for insolvency. It is a long-established financial institution that specializes in providing banking and financial services to companies in the sector.It has a strong reputation and has never faced such regulatory action.' is.

in Software,   Web Service, Posted by log1h_ik