Does news predict the next day?
Yesterday's coverage tone against today's price direction, across
17 instruments and 2196 paired sessions.
No conditional pattern survives correction. Across every instrument and every news bucket, the follow-through rate after bullish or bearish coverage is indistinguishable from the base rate once multiple testing is accounted for. Coverage did not predict the next session's direction in this sample.
| Instrument | Base rate
P(up) any day |
After BULLISH news
P(up next session) |
After BEARISH news
P(up next session) |
After NEUTRAL news
P(up next session) |
| Apple |
56% |
58%[32, 81]n=12+2pt | 100%[21, 100]n=1+44pt | 56%[47, 65]n=118-1pt |
| Aussie / US Dollar |
48% |
52%[35, 67]n=33+3pt | 46%[29, 65]n=26-2pt | 47%[37, 58]n=78-1pt |
| DAX 40 |
52% |
42%[24, 61]n=24-10pt | 50%[28, 72]n=16-2pt | 55%[44, 65]n=88+3pt |
| Euro / US Dollar |
48% |
78%[45, 94]n=9+30pt | 0%[0, 79]n=1-48pt | 46%[38, 55]n=126-2pt |
| Sterling / US Dollar |
48% |
57%[37, 74]n=23+9pt | 60%[31, 83]n=10+12pt | 45%[35, 54]n=103-3pt |
| Microsoft |
52% |
100%[21, 100]n=1+48pt | n/a | 52%[43, 60]n=130-0pt |
| Nikkei 225 |
54% |
49%[35, 63]n=45-5pt | 54%[35, 72]n=24+0pt | 58%[45, 70]n=55+4pt |
| Nasdaq 100 |
56% |
57%[39, 73]n=30+1pt | 50%[24, 76]n=10-6pt | 56%[46, 66]n=91+0pt |
| NVIDIA |
52% |
n/a | n/a | 52%[43, 60]n=131+0pt |
| S&P 500 |
53% |
32%[15, 54]n=19-22pt | 0%[0, 79]n=1-53pt | 58%[48, 66]n=111+4pt |
| Tesla |
51% |
n/a | 20%[4, 62]n=5-31pt | 52%[44, 61]n=126+1pt |
| US Dollar / Loonie |
56% |
54%[39, 68]n=39-2pt | 59%[36, 78]n=17+3pt | 57%[45, 68]n=69+1pt |
| US Dollar / Franc |
62% |
59%[39, 77]n=22-3pt | 62%[31, 86]n=8+0pt | 63%[51, 74]n=63+1pt |
| US Dollar / Yen |
64% |
58%[32, 81]n=12-5pt | 43%[16, 75]n=7-21pt | 66%[57, 73]n=119+2pt |
| WTI Crude |
54% |
33%[12, 65]n=9-21pt | 57%[25, 84]n=7+3pt | 56%[47, 64]n=115+1pt |
| Silver |
53% |
64%[39, 84]n=14+12pt | 42%[19, 68]n=12-11pt | 52%[43, 62]n=105-0pt |
| Gold |
48% |
40%[12, 77]n=5-8pt | 45%[21, 72]n=11-3pt | 49%[40, 58]n=115+1pt |
Why the base rate column is the whole point
A rising market makes every bullish reading look prophetic. If price rose 58% of all
sessions anyway, then 62% after bullish coverage is a 4-point edge, not a 62% hit
rate. And with ~40 observations that edge is well inside the noise.
The edge figure under each probability is the conditional rate minus the base
rate. It is the only number in the table that carries information. A bright green
"+18pt" with a narrow interval would be a real finding; a "+3pt" is not.
Small samples, wide intervals
The bracket under each percentage is a 95% Wilson interval. With 30, 40 observations
per bucket these are wide by construction, expect spans of 20 points or more.
Any bucket whose interval comfortably contains the base rate should be read as
undetermined, not as a small edge.
Tests are corrected with Benjamini, Hochberg across every instrument and bucket
examined. With this many comparisons, roughly one would clear an uncorrected 5%
threshold by chance alone; "survives" means it beat the corrected bar.
What this is not
Six months of one market regime, reconstructed after the fact from an archive that
thins with age, with a lexicon that assigns a crude direction to a headline. These are
associations, not causes; nothing here says coverage moves price. No forecasts, no
recommendations, no advice. If a pattern did appear, it would still be a description
of the past.
Open an Axi Account →
Descriptive statistics over published news coverage and settled price moves.
Not investment advice, not a recommendation, and not an assertion about any issuer's
conduct. Past association does not persist.
