Opening

Hi everyone, this is "AI Stock Market Shrimp Talk," and I'm Mu-Yan. Today is Tuesday, November 3, 2026, and we're forecasting the TAIEX for the next trading day, 2026-11-04 (Wednesday). Here's the conclusion up front: the model's direction for 11/04 is bearish, and the five-day short-term view points the same way.

This episode offers no buy or sell recommendations on individual stocks. The stock radar section later is purely a technical description of strategy screening results. Also, there are a few data quality issues in this episode that I'll lay out openly, including one "pending verification" percentage-change field, so please listen through and judge for yourself.

Today's Index Forecast

The service forecasts only the TAIEX, across two horizons: "next day" and "short term." Next day = 2026-11-04 (Wednesday); short term = roughly one week (5 trading days) starting 11/04.

*Confidence values are uncalibrated: over the last 19 "next day" forecasts, actual directional accuracy was only 47.4%, so please don't read 75% as a win rate. Timestamp for the current price of 45211.83: should be the 2026-11-03 close. Next-day percentage change: pending verification. The -2.49% figure from the data source appears simultaneously in the "next day" field, the "short term" field, and an interpretation report dated 10/30, suggesting it's the same stale, un-updated field. To avoid passing along an incorrect number, this episode will not cite that percentage. Please go by the target range of 44383–45046 instead.

In plain terms, the model sees 11/04 as weak. The entire next-day target range sits below the current price, and the upper bound of 45046 is still some distance below spot, meaning the model hasn't left much room even for a "chop around flat" scenario. The five-day short-term range is lower and wider, which says the model expects volatility to expand over the week rather than a single small down day and done.

That said, I have to be honest: this bearish conclusion rests largely on external market pressure (you'll see the SOX collapse in the next section), while the latest internal structural data for Taiwan's market is missing from this dataset.

Global Market Snapshot

What follows is current status only, with no forecasts.

Timestamp caveat: The data source lists the TXF night session at 44433 but doesn't specify which session it corresponds to. If Taiwan's market traded normally on 11/03, this is more likely the 11/02 night session close, not the post-11/03 night session. 11/03 post-close night session data is still pending. The four major US indices and the SOX are likewise reference points ahead of Taiwan's 11/04 open, and do not represent the 11/04 forecast.

The most glaring item in this table is the SOX at -7.2%. The Nasdaq itself fell 2.86%, but semiconductors dropped 7%, meaning the selling wasn't broad-market, it was precisely targeted at the AI and chip complex. Taiwan's index weighting is heavily concentrated in semiconductors, and that's the main external basis for the model's bearish call on 11/04.

Strategy Stock Radar

The following are strategy screening results, not individual stock forecasts. This program does not produce directions, target prices, or confidence values for any individual stock. Below are only screening scores, valuation tags, and technical indicator descriptions.

Technical signal strength (quantified signal score, scale -3 to +3; this is a separate metric from the 100-point "composite score" above, so don't mix them). The following is purely technical description, not investment advice:

• Yang Ming (2609): Signal score +1.0, leaning bullish. A bullish engulfing candlestick pattern appeared. P/B ratio 0.68.
• Formosa Petrochemical (6505): Signal score +1.2, leaning bullish. Williams %R at -95.94, in oversold territory, a technical bounce-after-selloff pattern description. P/B ratio 1.92.
• Hon Hai (2317): Signal score 0.0, neutral. Moving averages tangled with no clear direction; price broke below the monthly MA (medium-term weakness); RSI 48; KD turning down short term. P/B ratio 1.80.
• Novatek (3034): Signal score 0.0, neutral. Short-, medium-, and long-term MAs all rising (bullish MA alignment), price above the monthly MA (medium-term bullish); RSI 54; KD turning down short term. P/E and P/B data unavailable.
• Chunghwa Telecom (2412): Signal score -1.7, leaning bearish. CCI at 122.18 shows strengthening momentum, but RSI 83 and Williams %R at 0 both sit in overbought territory. P/B ratio 3.39.

A few things worth calling out. First, four names share an institutional flow score of 100. The data source provides no definition of this score's upper bound or calculation basis, and doesn't say whether maxing out is routine, so please don't read "100" as "extremely strong institutional support." I can't vouch for that interpretation. Second, Yang Ming has a P/B of just 0.68 yet carries an "expensive" tag, while Formosa Petrochemical at 17x earnings is tagged "cheap." This means the valuation tag compares against peers or the stock's own historical range, not absolute levels, so don't read the two together. Third, the highest signal score among these five is only +1.2. The group is broadly neutral, with no strong signals anywhere.

Terminology Notes

So episodes can be compared against each other, let's define the terms used in the next section:

• Directional accuracy: The proportion of forecasts where the predicted direction matched the actual closing direction.
• Full range hit: Whether the trading day's actual close fell within the predicted target range. (Note: the data source doesn't specify whether this is judged on the closing price alone or requires the day's high and low to both fall inside the range. We use the most lenient closing-price interpretation here.)
• Average error: The average percentage deviation between predicted and actual values. (The data source doesn't specify whether this is computed from the range midpoint or a range boundary; definition pending.)

Recent Forecast Accuracy Review

Here are the numbers straight, no spin. Index, next day: 9 correct directions out of the last 19, directional accuracy 47.4%, full range hits 5/19, average error 1.17%. Index, short term (5 days): 8 correct out of the last 15, accuracy 53.3%, full range hits 3/15, average error 2.51%.

That next-day figure of 47.4% is worse than a coin flip. So today's 75% confidence value in the table is badly disconnected from actual performance, which is exactly why I added the footnote directly in the table. I don't want listeners who only skim the table to be misled by that number. Short term at 53.3% is marginally better, but 15 samples is far too small to be statistically meaningful.

One more pass on this episode's data quality: the percentage-change field is pending verification, the night session timestamp is ambiguous, and the data source hasn't updated 11/03 market breadth (advancers/decliners). The only breadth figure available is from 2026-10-30, 92 advancers against 952 decliners. That's last Friday's number, with 11/02 and 11/03 both having traded since, so it can't be treated as the current market structure. I'd rather name these gaps than force together a narrative that merely looks complete.

Disclaimer

This content is automatically generated by an AI model from quantitative data and is for research reference only. It does not constitute investment advice or any recommendation. This site and the program's producers are not a licensed securities investment consulting business, do not provide analytical opinions to specific individuals, and receive no compensation for this content. Investing carries risk; please exercise your own judgment and bear your own responsibility.