Analyst’s summary
Ming-Chi Kuo is an analyst at TF International Securities, known for research on Apple and its supply chain. He reaches his audience through posts on X. Across the record, most of his claims are mentions of an asset with no direction attached; the minority that carry a direction are all single equities, almost all about business outcomes, and none names a price level. His most-called names are AAPL, TSM and NVDA.
How accurate is he?
He is accurate 36.7% of the time against a base rate of 41.2% for the stocks he calls, so he scores 4.5 points below it. Accurate means he hits or partially hits the hurdle in the direction his signal implies, inside the window he named or the calculated window determined from our algorithm; for a price target it means the target was reached inside the window, or half the stated move was.
How does he compare to just holding the S&P 500?
If you had put $1 into the S&P 500 on the day of his first equity call in April 2022 and left it there until his last equity call was settled, you would have $1.68. If you had put $1 into every one of his 189 equity calls instead, buying when he posted and selling when each call was due, you would have $1.44; the S&P 500 held on those same days made $1.66.
Taking profit early, at twice the hurdle with no stop-loss, turned $1 into $1.10. The base rate on the stocks he picks is $2.17: the same stocks, bought on random dates in the same period, held for the same lengths of time, with the same long and short sides.
Explainer on his accuracy against his return
Accuracy counts how often he was right; return counts how far the price went. He was right less often than chance, and a dollar compounds the path, not the average: his hits averaged +31.5% and his misses -25.2%, and two positions lost more than their whole value. (* see the footnote below for a fuller explanation)
How does following his timing compare to the base rate of the stocks he chose?
A dollar following every one of his calls to the end of its window became $1.44 against a base rate of $2.17; with a stop at the hurdle and a take-profit at twice it, $0.65. In one of his five years he beat the base rate, in 2023; over the last twelve months he scored 35.9% against a base rate of 46.2% on 29 calls.
What is the best holding period if we follow Kuo?
One day: a dollar following every call for one day became $1.45 against a base rate of $0.88, and no other plain hold does better. On tracking, his edge over the stocks is largest at 365 days, +15.3% against +14.4%, and at ninety days he is 1.5 points behind them.
Does he have any names where he has a particular edge?
On 2382.HK, 005930.KS, 2317.TW his calls beat the stock's own 30-day move; on AAPL, the name he calls most, they beat it by 0.4 points; they trailed it on 002241 and QCOM.
In our opinion, this is a voice to read for information about Apple's supply chain, not a timing signal. Following his calls as a whole did no better than buying the same assets on random dates, and longer holds trailed the stocks' own drift. The exceptions are narrow: his 30-day calls beat the stock's own move on 2317.TW, 005930.KS and 2382.HK, and barely on AAPL, while trailing on QCOM and 002241. A follower should treat the broad record as context and consider acting only where the edge is measured.
Notes
1 Right means the hurdle test. Every stock carries a hurdle, T, that depends on its class and on how long the call was given (the table on page 2): 5 percent for a single equity at ninety days, under half a percent at next close, eleven at a year; 13 percent for crypto at ninety days. A full hit is a move of at least twice T in the direction he called by the end of the window and counts 1.0; a partial hit is at least T and counts 0.8; a miss, and a flat outcome that moved less than T either way, count zero. A price-target call is a full hit when the target is reached inside the window, a partial at half the stated move and a miss otherwise; it has one reading, the touch, so a later fall does not change it and the hit rate is not a holding return. Every resolved call is scored and every one is in the denominator. At expiry reads the price once at the window's end; one-touch counts the same move if it happened at any point inside the window.
2 Base rate: for each call, over every window of the same length on the same stock across the record, from two years before his first call to the last price and leaving out the call's own window, the share that cleared the same hurdle in his direction, with a partial at 0.8 and everything else zero; built exactly as the hit rate is, then averaged over his calls. Lift is his rate minus it. The percentiles and the drift in Parts 2 and 3 use a narrower set, the 730 days before each call.
3 The percentile is where a call's return ranks among all windows of the same length in that stock's own past two years; the 50th is chance. The placebo band is what random posting dates on the same stocks produce; a reading inside it is indistinguishable from chance.
4 The dollar base rate: the same stocks, the same number of calls, the same long and short sides, bought on random dates; the median of 150 such runs. Each call is held to its own window, and a price-target call is sold at its target if that comes first, on his dates and on the random ones alike.
5 Win rate and return per call: a follower buying at the first close after each post and selling thirty days later, costs deducted; the share of calls that made money, and the average result.
6 Drawdown: the deepest fall of that follower's account from its peak.
7 A bracket is a take-profit and a stop-loss set in advance; here both are set in units of the call's own hurdle, and a price-target call takes its target as the take-profit, so a take-profit fill is a hit banked and a stop fill is a miss taken. It is a rule of ours, not part of his record, and it never feeds the alpha gate.
* Accuracy counts calls. Every call counts once, whatever the size of the move: a full hit is one, a partial is 0.8, a miss or a flat is zero. Fifty-five of his 189 calls were full hits, eighteen partial, 116 were misses or flats. Random windows on the same stocks clear the hurdle more often than that, because stocks of this kind move the hurdle's distance inside the window about 33.5% of the time on their own. So by count he is right less often than chance. Return weighs calls by how far they moved. His fifty-five hits averaged +31.5%, one of them, TSM, +95.0%. His sixty-four misses averaged -25.2%, and his fifty-two flats were small. Add them up and the winners are worth +2062 points against -1713 for the losers. A dollar compounds the path, not the average: two of his positions, shorts held through a rally, lost more than their whole value, and with no stop-loss they took the account down whatever the other calls did; 2.14b shows the same calls with a stop.
Summary of metrics
Integrity: 1066 posts were read, yielding 932 claims, 210 directional calls and 189 resolved. (see section: the worked claim, Part 5, annex)
Information: scoring for accuracy by our rules, he is accurate 36.7% of the time, against a base rate of 41.2% for the same stocks, so he scores 4.5 points below it. On tracking his calls, his best horizon is 365 days, where his calls return +15.3% against +14.4%, an edge of 0.9 points. At the ninety-day horizon his calls return +2.7% on average, 1.5 points behind the stocks' own return of +4.1% over the same days. (see section: 1.3, 1.18, 2.19)
Robustness: in 1 of his 5 years he scored above the base rate, in 2023; over the last twelve months he scored 35.9% against a base rate of 46.2% on 29 calls; the record holds 189 resolved calls on 75 stocks. (see section: 1.6, 1.9, 5.3, 5.4)
Concentration: his calls beat the stock's own 30-day move on 2382.HK, 005930.KS, 2317.TW; on AAPL, the name he calls most, they beat it by 0.4 points; they trailed it on 002241 and QCOM. (see section: 2.8)
Response profile: if we follow all of Kuo's calls, the holding period most profitable is one day. One dollar following every call for one day became $1.45 against a base rate of $0.88. There is no benefit to holding his calls for 7 days, 30 days or each call's own window. (see section: 1.3, 2.13, 2.14b)
Distribution: one call is worth +1.7% on average, after costs; the middle half of his calls end between -10.2% and +15.0%, and 52.9% of them end positive. Following every call to the end of its window turned one dollar into $1.44 against a base rate of $2.17, and the account fell 26.1% from its peak along the way. (see section: 2.13 to 2.16)
accuracy at expiry (vs base rate at expiry of 41.2%)
accuracy on one-touch (vs base rate one-touch of 69.1%)
one dollar following all of his calls vs one dollar holding the index for the same period
005930.KS +2.9%/-0.8%
2317.TW +4.2%/+0.5%
the names where his edge is biggest: his average 30-day return / the stock's own
48.7% / +0.4% (30 day)
49.2% / +0.2% (7 day)
53.4% / +0.4% (1 day)
win rate / return per call, by holding period; aggregate is each call held to its own window
$0.88 vs base rate $1.61 (30 day)
$1.16 vs base rate $1.02 (7 day)
$1.45 vs base rate $0.88 (1 day)
one dollar following every call vs the base rate for following the same stocks on random dates
selling each call as soon as it is up twice its hurdle, no stop-loss: 66.1% of calls win against 52.9% on a plain hold to the window, but one dollar ends at $1.10 against $1.44: capping the winners costs more than it saves
+6.8% vs +8.4% (180d)
his average return against the stocks' own, at the horizons he uses most (ten or more calls); the difference is his edge in points
Following Ming-Chi Kuo's calls from the first close at or after each post and holding each call to the end of its own window, one dollar became $1.44 over 1643 days in market, against just holding the S&P 500 for his equity calls ($1.44 following those calls against $1.68 held), and against holding the same stocks Kuo chooses on random dates, which would have produced $2.17.
This report does not establish an investment advantage. Comparison data for similar assets was unavailable.
Bible v0.46 · PRD draft-81 · Ruling Ledger v85 · effect-2.0, reduced to own-history in this run · Specimen v12.15