Challenging Japanese Stock Swing Trading with ChatGPT #30-31
I continue to engage in short-term swing trading, where I purchase Japanese stocks during my weekday lunch break and, in principle, sell them during the lunch break of the next business day. I use ChatGPT for stock selection and trading decisions, verifying with my own money whether it is actually possible to make a profit in Japanese stock trading using AI.
This article is also written by ChatGPT, with the exception of the final section, “My Thoughts This Time.” The main text is organized based on actual trading records and the analysis performed at that time.
Three consecutive losses totaling -34,830 yen, followed by +31,650 yen the next day. Over two days, it is almost back to square one at -3,180 yen.
This time, I will summarize the two days of October 8th and October 9th.
On October 8th, I cut losses on all three stocks I had purchased the previous day—Shin-Etsu Chemical, TDK, and AGC—resulting in -34,830 yen. On the other hand, the following day, October 9th, JT, Fujitsu, and Oriental Land turned a profit, allowing me to recover to a total of **+31,650 yen** across five stocks.
In other words, the realized profit/loss for the two days is **-3,180 yen**. While it looks like I recovered almost the entire amount the day after a significant loss, looking at these two days in detail reveals quite clearly the weaknesses in stock selection by ChatGPT and the points that need improvement.
October 8th: Shin-Etsu Chemical, TDK, and AGC all ended in the red.
On October 7th, I purchased 100 shares of Shin-Etsu Chemical at 6,264 yen, 100 shares of TDK at 3,561.9 yen, and 100 shares of AGC at 5,936.4 yen. At noon on the following day, October 8th, I sold Shin-Etsu Chemical at 6,153 yen for **-11,100 yen**, TDK at 3,406 yen for **-15,590 yen**, and AGC at 5,855 yen for **-8,140 yen**. The total for the three stocks was -34,830 yen**.
TDK was particularly painful, falling about 4.4% from the purchase price. Shin-Etsu Chemical also fell about 1.8%, and AGC fell about 1.4%, with all three stocks moving in the same direction.
I need to evaluate ChatGPT’s judgment from the previous day quite strictly here. It wasn’t just a misreading of the magnitude of the rise; it got the direction itself wrong for all three stocks by the following noon.
Furthermore, the problem lies in the combination of Shin-Etsu Chemical, TDK, and AGC. Although the industries themselves are different, they were stocks that are easily influenced by the external environment of growth stocks, such as semiconductors, electronic components, and electronic materials. Even though I thought I bought three individual stocks, if the market collapses, I am effectively holding three similar factors.
This result was an easy-to-understand failure example showing that “increasing the number of stocks does not necessarily mean diversification.”
October 9th: 3 wins and 2 losses. JT alone accounted for +26,900 yen.
At noon on October 8th, I significantly changed my positions.
I purchased 100 shares of Japan Tobacco (JT) at 6,978 yen, 200 shares of Sumitomo Chemical at 578.6 yen, 100 shares of Mitsubishi UFJ Financial Group at 3,524 yen, 100 shares of Fujitsu at 4,142 yen, and 100 shares of Oriental Land at 2,917.5 yen.
The results of the sales at noon on October 9th were: JT at 7,247 yen for **+26,900 yen**, Sumitomo Chemical at 569.6 yen for **-1,800 yen**, Mitsubishi UFJ at 3,459 yen for **-6,500 yen**, Fujitsu at 4,220 yen for **+7,800 yen**, and Oriental Land at 2,970 yen for **+5,250 yen**. With 3 wins and 2 losses across 5 stocks, the total was **+31,650 yen**.
Of this +31,650 yen, JT alone earned +26,900 yen. That is about 85% of the total profit. On the other hand, Mitsubishi UFJ was -6,500 yen, so it is not the case that I correctly read the direction for all stocks.
Even so, compared to the three losses on October 8th, the following day was clearly an improvement. Because stocks with different price movements—JT, Fujitsu, and Oriental Land—made a profit, the portfolio was not structured in a way where all stocks sank together the moment one market factor collapsed, as it had the previous day.
How accurate was the “previous day’s forecast”?
Lining up the two days, the characteristics of short-term predictions by AI become visible.
October 8th was 0 wins and 3 losses, and the prediction of direction itself failed. Moreover, because the losses occurred simultaneously, there was a problem not only with individual stock selection but also with correlation management of the entire portfolio.
In contrast, October 9th was 3 wins and 2 losses, resulting in a profit. However, it was highly dependent on the significant rise of JT, and it is not a result that indicates ‘all 5 stocks were predicted with high accuracy’.
What is important from this difference is not just looking at the win rate, but looking at how many stocks fail simultaneously when you get it wrong.
The quality of risk is quite different between a configuration where you buy 3 stocks and all 3 fall due to the same external factor, and a configuration where you can absorb the loss even if 2 out of 5 stocks fail with the profits from the remaining ones, even if the AI selection is the same.
In fact, as of October 8th, the cumulative profit/loss for October had worsened to -58,470 yen, and the total cumulative to -86,830 yen. However, thanks to the +31,650 yen on October 9th, the October cumulative has recovered to -26,820 yen, and the total to -55,180 yen.
On October 9th, 4 stocks were newly purchased. Oriental Land was entered again.
At noon on October 9th, I newly purchased 100 shares of KOKUSAI ELECTRIC at 9,921 yen, 100 shares of Seven & i Holdings at 2,006.5 yen, 100 shares of Oriental Land at 2,964 yen, and 100 shares of Chubu Electric Power at 2,633.8 yen. The total purchase amount is 1,752,530 yen.
Chubu Electric Power was 2,646 yen at 11:30 AM, with a VWAP of approximately 2,635.9 yen, but the actual purchase price was 2,633.8 yen. At least looking only at the execution price, it is not a form of chasing the morning session’s rise at a high price.
Also, while incorporating the semiconductor manufacturing equipment stock KOKUSAI ELECTRIC, I am also holding Seven & i, Oriental Land, and Chubu Electric Power at the same time, making it a configuration with more diversified factors than the 3 stocks on October 7th.
What has become clear through this verification
The biggest thing this time is not that the -34,830 yen on October 8th and the +31,650 yen on October 9th almost offset each other.
What is important is the content.
On October 8th, all 3 stocks were negative, and rather than ‘individual stocks chosen by AI’, it was a state of betting collectively on one market environment. On the other hand, October 9th secured a profit despite being 3 wins and 2 losses. However, most of the profit was due to one stock, JT, and it is still too early to conclude that the reproducibility of stock selection has been confirmed.
In the performance ledger, up to the settlement on October 9th, there have been 83 confirmed trades, 42 wins and 39 losses for the 81 cases where individual wins and losses can be determined, and a total confirmed profit/loss of -55,180 yen. The win rate itself is over 50%, but the cumulative profit/loss is still negative.
What I want to emphasize more in this verification in the future is not ‘how many stocks I got right’, but how much I can minimize losses when I get it wrong, and whether multiple stocks will not lose at the same time due to the same factor.
Note that the 4 stocks purchased on October 9th will be held over the long weekend, not the usual next business day. Since the Tokyo Stock Exchange is closed on October 12th, the planned exit is during the lunch break on October 13th. Because the holding time is longer than usual, overnight risk due to overseas markets and news over the weekend will also be a new verification material this time.
My thoughts this time
・I worked overtime for the first time in a while, so I’m summarizing two days’ worth of articles. ・On the 7th, I believed in semiconductors, so I went all-in on related sectors, and suffered a quite painful defeat. ・On the 8th, I thought, ‘My funds are just being whittled away, I’ve had enough,’ so I bought JT and Mitsubishi UFJ with long-term holding in mind, but JT rose significantly. I can play a little longer now. ・JT’s overseas tobacco stocks are doing great, and it seems there was an article in the Nikkei saying JT is doing well too. I was just lucky. ・Today, Rorze’s financial results were good and it skyrocketed. Chappy said there is a possibility that it will spread to semiconductor manufacturing equipment, so I bought KOKUSAI ELECTRIC, which was at an affordable price. At the closing price, it had risen by more than 10,000 yen, so please, just like this…
※This article is a record of personal trades and verification, and does not recommend buying or selling specific stocks.
#Investment
#StockInvestment
#JapaneseStocks
#Stocks
#AssetManagement
#AssetFormation
#SwingTrading
#InvestmentRecord
#IndividualInvestor
#StockInvestmentBeginner
#ChatGPT
#AIInvestment
#Trading
#InvestmentBeginner