The US and Japan raised interest rates in the same week, yet the yen weakened. Here is how the AI investment team interpreted the reason.
In September, the central banks of the United States and Japan raised interest rates in the same week.
According to the textbook, if Japanese interest rates rise, the yen should be bought, leading to a stronger yen. However, in reality, the yen weakened further.
How did the AI investment team, “AI GOLDMAN,” interpret this discrepancy? The focus this time is not on the team’s performance, but on the team’s “thought process.”
First, four premises:
・AI GOLDMAN is an AI investment team I created using Claude Code. The starting capital is a virtual 1.5 million yen, and no actual money is being moved. The figures for the team’s assets are listed at the end of the article.
・The analysis is done by the AI team, but the final decision is made by me. The AI is not trading automatically.
・I will not write down specific stock names or the timing of trades.
・What is written here is a record of “how the AI team thought.” It is not a promise that the predictions will be correct, nor is it a recommendation to buy or sell anything. The outlook and probability figures are merely the team’s assessment at the time of recording.
1. First, look at how the world is moving
The team’s analysis always begins in the same order. The first to speak is Victor (Macro Strategist), who is in charge of looking at the global economy. Talk of individual assets comes after that.
The facts that Victor (Macro Strategist) laid out this week were as follows.
・United States: Interest rate hike on September 16. Policy rate is now 3.75–4.00%. 16 out of 18 participants expect “further hikes”.
・Japan: Interest rate hike on September 18. Policy rate is around 1.25%. A rate hike just three months after the previous one.
・Exchange rate: The yen, which had strengthened to the 153 yen per dollar range in mid-September, returned to around 158 yen per dollar on September 24 (as of September 24, 2026).
Even though both countries raised interest rates, the yen weakened. Victor (Macro Strategist) provided three explanations for this.
━━━━━━━━━━━━
🏦 AI GOLDMAN
Investment Committee Meeting
Agenda: Why did the yen weaken despite the interest rate hikes?
━━━━━━━━━━━━
🌎 Victor (Macro Strategist)
“First, let’s talk about how the world is moving. (1) The US interest rate hike was perceived as something that ‘will continue,’ leading to the view that the interest rate gap between the US and Japan will not narrow. (2) Japan’s interest rate hike was sufficiently priced in beforehand, and there is a possibility that the yen was sold as a ‘sell the fact’ move the moment it was decided. (3) The 158 yen per dollar level is being watched as a level where government intervention to buy yen is feared.”
🌎 Victor (Macro Strategist)
“However, (2) is an estimation, and it has not been confirmed whether the intervention in (3) actually occurred. I will write that only (1) could be confirmed as a fact.”
━━━━━━━━━━━━
A characteristic of this team is well-represented in the second statement. After providing three explanations, it sorts them itself by “what could be verified.” In the report, labels such as “Fact,” “View,” “Estimation,” and “Unknown” are attached to almost every line.
2. If a premise is wrong, discard the premise
In fact, as of August, the team had set its exchange rate monitoring line based on the premise that “if Japan raises interest rates, the yen will strengthen.”
That premise was proven wrong by the Bank of Japan’s interest rate hike in September. Robert’s (Risk Manager) judgment at that time was as follows.
━━━━━━━━━━━━
🛡️ Robert (Risk Manager)
“The idea that ‘interest rate hike leads to a stronger yen’ has been partially disproven by this result. However, that does not mean the risk of a stronger yen has disappeared. We should consider that the risk has gone from one-directional to two-directional.”
━━━━━━━━━━━━
It does not say, “It was wrong, but it must have been a fluke.” It rewrites the premise itself. The line that was only monitoring for a stronger yen was redesigned to react to both a weaker yen and a stronger yen.
You cannot prevent a prediction from being wrong. What you can prevent is continuing to use a prediction that has already been proven wrong.
3. Giving a ‘name’ to the current market
This week, Victor (macro strategist) gave a name to the current state of the market: ‘Risk-on under hawkish coordination’.
Breaking down the terms, it goes like this:
・Hawkish coordination: Both the US and Japan are on the side of raising interest rates together. ・Risk-on: Even so, stocks are in a high price range (Nikkei Stock Average is in the 65,000 yen range as of September 24, 2026).
When interest rates rise, it is usually a headwind for stocks. The reason stocks are still rising is that the market sees ‘the growth in corporate earnings as outpacing the rise in interest rates.’ That is how the team interpreted it.
On the other hand, there are assets that are falling in price, pushed down by the same interest rate hikes.
・Gold: Since it is an asset that does not generate interest, it tends to be at a disadvantage when interest rates rise. It is around $4,290 per ounce (as of September 23, 2026), and has been undergoing a correction since its August high. However, the team’s view is that tensions in the Middle East are providing a floor. ・Crypto assets: King’s (crypto assets) judgment is ‘no clear direction.’ Buying and selling factors are playing tug-of-war, and it is moving within a certain range (as of September 24, 2026).
Once it has a name, it becomes possible to compare. Will it keep the same name next week, or will it change? If it changes, that is a week where the team should review their premises. In this series, we will follow the changes in this name.
4. Writing ‘where we might be wrong’ before the forecast
There is a column in this team’s report that I find most interesting. It is the ‘conditions under which we could be wrong’ column, placed right after the outlook. This week, Victor (macro strategist) wrote three things.
-
Crude oil prices surge due to tensions in the Middle East, reigniting inflation. If that happens, both the US and Japan may raise interest rates faster than planned, and stocks bought on high expectations could potentially collapse.
-
US interest rate hikes are perceived not as ‘good hikes because the economy is strong,’ but as ‘bad hikes because inflation cannot be stopped.’ At that point, stocks and crypto assets could fall simultaneously.
-
The sudden rise in semiconductor and AI-related stocks could potentially be unwound in a reaction.
Furthermore, another column lists ‘watchdog’ indicators, such as ‘what would be a sign if it happened.’ These include a sharp rise in crude oil prices, a large one-day drop in semiconductor stock indices, and the exchange rate approaching past intervention levels.
Another thing that left an impression was how they read the ‘VIX index,’ which is said to reflect market anxiety. The value at the beginning of September was 16.34, which is in the normal range (as of September 1, 2026). They did not read this as a ‘reassuring factor,’ but wrote that it ‘might be a sign that optimism is going too far.’ On the other hand, they also wrote that ‘this time it is accompanied by actual demand, so the nature is different from past bubbles,’ and concluded by calling it a ‘point of contention where judgments are divided.’
They do not commit to one side or the other. If it is divided, they write that it is divided.
5. Always attach ‘how much confidence you have’ to a forecast
When writing an outlook, the team always divides scenarios into multiple parts and assigns a probability to each.
James’s (bonds/interest rates) outlook for Japanese interest rates was as follows:
・Main (60% probability): The Bank of Japan continues to raise interest rates, and the yield on 10-year government bonds gradually rises to the low 3% range. ・Alternative (25% probability): Inflation concerns intensify due to the Middle East situation, and the pace of interest rate hikes accelerates. ・Risk (15% probability): Stock prices fall starting from the US, government bonds are bought as safe assets, and yields temporarily fall.
The yield on the 10-year government bond was 2.95–2.985% between September 18 and 24. This is the highest level since the mid-1990s (values based on multiple search results; confirmation with official Ministry of Finance figures is deferred to next time).
King (Crypto Assets) also divided the year-end outlook for crypto assets into three categories: 25% bullish, 50% base, and 25% bearish. Furthermore, he explicitly added that these percentages are ‘not strict probabilities.’ It is meant as a gauge of how confident he is.
King’s (Crypto Assets) report also has another section called ‘Information unfavorable to my own theory.’ What was written there this week was: ‘Approximately $1 billion flowed into US spot crypto asset ETFs on September 21 alone. This is evidence of actual demand and indicates the possibility that my cautious view may be wrong.’
Searching for and writing down information that is inconvenient for one’s own opinion. I think this is something even human analysts find difficult to do.
6. Write ‘unknown’ for things you don’t know
Victor’s (Macro Strategist) report contains a list of ‘unknowns’ in almost every issue. This week there were six. These include the latest US inflation figures, the current price of crude oil, and how much of a rate hike is priced in for the next meeting.
For example, crude oil is one of the most important figures related to the first of the ‘conditions under which I could be wrong.’ He writes down without hiding the fact that he hasn’t obtained it, and carries it over to next time.
The same attitude is evident in how numbers are handled. This week, there was a discrepancy between the figures assembled from search results and the figures taken directly from a stock price website. Robert (Risk Management Officer) decided ‘not to use’ the assembled figures and reviewed them using only the figures taken directly. He chooses figures with a clear source over plausible-looking ones.
Thoughts after watching from the sidelines for 5 weeks
If you ask an AI, ‘What will the market do next week?’, it will return any number of plausible-sounding answers.
But what became visible by forming a team was not the answer itself, but the ‘method’ of arriving at the answer.
・Write facts and estimates separately ・Discard premises if they prove incorrect ・Write where you might be wrong before making a prediction ・Add a degree of confidence with numbers ・Write down inconvenient materials and unknowns yourself
No one knows whether the market will go up or down. Even so, with the assessment of a team that continues these five things every week, you at least know ‘how far you can trust it.’ That was probably what I wanted most from AI analysis.
Next time
Next time, I will also write about how the AI team read the market that week. This will include a review of how this time’s ‘conditions under which I could be wrong’ turned out in one week.
■ Today’s figures (as of September 24, 2026 / virtual funds) ・Total initial capital of 1.5 million yen: +2.9% since the start of operations (August 19, 2026) (no previous calculation) ・Domestic stocks: approx. +3% ・Gold: approx. -2% ・Crypto assets: approx. +30% ・Assets that reached the exit line: None ・Asset movement (since the last posting, executed trades only): September 24, sold a portion of crypto assets (profit taking) (assumed transaction on the ledger) ※ Domestic stocks and gold are calculated based on the closing price of the most recent trading day, and crypto assets are calculated based on the price at the time of aggregation. This is a record using virtual funds; the operation period is short, and the figures are heavily influenced by luck. I post them in the same format whether the day was up or down.
📚 You can follow this series in the magazine ‘AI GOLDMAN’
⚠️ About this article
・AI GOLDMAN is a fictional AI investment team built on Claude Code. It has no relationship whatsoever with the real Goldman Sachs or any other financial institution. ・This article is a record of an AI analysis experiment and is not investment advice. It does not recommend the buying or selling of specific stocks or products, nor does it guarantee investment results. ・The figures and analysis in this article may contain errors. Please make final investment decisions at your own risk. ・The outlook and confidence levels in the article are records of the analysis provided by the AI team at the time of writing and do not indicate future price movements or the team’s future trading plans. ・The profit/loss ratios and asset movements in this article are records of assumed transactions conducted with a virtual initial capital of 1.5 million yen (calculated mechanically at the price on the aggregation date). No actual money is moving. ・Because the operation period is short, the figures are heavily influenced by luck. They do not indicate future results. I post them in the same format whether the day was up or down. ・Individual stock names, amounts, quantities, prices, and future trading plans are intentionally omitted. It does not recommend the purchase or sale of specific assets, stocks, or products. ・Although the analysis is performed by the AI team, I (HIRO) make the final decision on the assumed transactions. The AI is not trading automatically.
Hashtag
#TriedWithAI #AIGOLDMAN #AIInvestmentTeam #ClaudeCode #AIAgent #MultiAgent #SubAgent #AIOrganization #Claude #Anthropic #GenerativeAI #AIUtilization #AIExperiment #ExperienceReport #AIInvestment #InvestingWithAI #AIAnalysis #MarketAnalysis #MarketTrendAnalysis #MarketAnalysis #EconomicNews #NewsCommentary #MonetaryPolicy #InterestRateHike #BOJ #FRB #FOMC #InterestRates #LongTermInterestRates #GovernmentBonds #ForeignExchange #YenDepreciation #YenAppreciation #DollarYen #GoldPrice #Gold #CryptoAssets #Cryptocurrency #NikkeiAverage #StockMarket #Semiconductors #CrudeOil #MiddleEastSituation #GeopoliticalRisk #VIX #ScenarioAnalysis #RiskManagement #Investment #InvestmentBeginner #InvestmentStudy #MoneyStudy #AssetManagement #AssetManagementBeginner #AssetFormation #MoneyLiteracy #ThinkingMethod #HypothesisThinking #FactCheck #CriticalThinking #TriedIt