A daily-return correlation study of BTCUSDT and ZECUSDT for September 2026, set against the 2025–2026 record, with ETHUSDT as a control. The suspected strong inverse relationship does not survive contact with returns.
Question. Do BTCUSDT and ZECUSDT candles show a strong opposite (negative) correlation in September 2026?
Answer. No. On daily returns — the statistically meaningful object — the September relationship is positive but weak: Pearson +0.34 (n = 21, p = 0.13), with a 95% confidence interval that spans zero, [−0.11, +0.67]. Thirteen of twenty-one days moved the same way. The only large negative number in the data is the full-history correlation of price levels (−0.60) — and that is a textbook spurious artifact: BTC fell −8.9% while ZEC rose +2,699% over the window, so two non-stationary trending series with opposite slopes correlate negatively. In September itself, when both rose, the level correlation is positive (+0.49).
There is no evidence of a strong inverse relationship between BTC and ZEC. Returns co-move weakly and positively; the negative coefficient lives entirely in the levels of two price series that drifted apart.
The working suspicion is that BTCUSDT and ZECUSDT candles moved inversely during September 2026 — a strong negative correlation. If true, it would be economically useful: a hedge, a pairs trade, or a rotation signal.
H. The daily returns (and perhaps the price paths) of BTC and ZEC are strongly negatively correlated in September 2026. Hypothesis under test
We test it directly, distinguish price levels from price changes, and place September inside the full 2025–2026 record so the month can be judged against its own history. ETHUSDT enters as a control asset that should, under any sensible crypto market-factor model, correlate positively with BTC — if it does not, the method is broken rather than the market.
Daily (1d) klines from Binance Public Data bulk archives, futures USD-M perpetual, normalized to canonical Parquet by the lab and validated cleanly. Three symbols are aligned on the UTC day by inner join: BTCUSDT, ZECUSDT and ETHUSDT.
| window | candles | returns | first | last | missing days |
|---|---|---|---|---|---|
| September 2026 | 22 | 21 | 2026-09-01 | 2026-09-22 | 0 (month incomplete) |
| Full history | 630 | 629 | 2025-01-01 | 2026-09-22 | 0 |
The stationarity contrast is decisive and is reported in Table 2: the price levels are indistinguishable from unit-root processes, while the returns are stationary.
| series | BTC p | ZEC p | ETH p | reading |
|---|---|---|---|---|
| log price levels | 0.58 | 1.00 | 0.46 | non-stationary |
| log returns | 0.00 | 0.00 | 0.00 | stationary |
Correlating non-stationary levels can produce arbitrarily large coefficients unrelated to any co-movement of changes. The levels result is shown for diagnosis, not as evidence of dependence.
The distinction is the whole story. Over the full 2025–2026 window the two assets moved in opposite directions at the trend level: BTC ended −8.9% (and spent much of 2026 falling) while ZEC compounded +2,699%. Two series with opposite drift trace out a large negative levels correlation.
Figure 1. Full-history indexed prices (log scale, start = 100). ZEC is an extreme idiosyncratic trend; BTC and ETH are flat-to-negative. Correlating these levels yields −0.60 for BTC–ZEC — a property of the slopes, not of daily co-movement.
| measure | September 2026 | Full history | comment |
|---|---|---|---|
| log levels, BTC–ZEC | +0.49 | −0.60 | sign flips with drift |
| log returns, BTC–ZEC | +0.34 | +0.44 | positive in both |
| log returns, BTC–ETH (control) | +0.93 | +0.85 | market factor intact |
In September both assets rose, so even their levels correlate positively (+0.49). There is no window in which a negative levels coefficient coincides with negative return co-movement — the negative number is a trend artifact that reverses sign the moment the drifts align.
ZEC nearly doubled in three weeks (+96.5%) while BTC rose +11.3% and ETH +13.9%. ZEC’s move is real but idiosyncratic, punctuated by double-digit daily swings in both directions.
Figure 2. September 2026 indexed prices (start = 100). Both assets rose; ZEC’s path is driven by idiosyncratic jumps (e.g. +15.5% on 03 Sep, +18.6% on 16 Sep) rather than by BTC.
| statistic | ZEC vs BTC | ETH vs BTC | reading |
|---|---|---|---|
| Pearson r | +0.342 | +0.930 | ZEC weak, ETH strong |
| Pearson p (2-sided) | 0.130 | <0.001 | ZEC not significant |
| Spearman ρ | +0.484 | +0.949 | rank agreement moderate |
| Kendall τ | +0.314 | — | — |
| R² | 0.117 | 0.865 | BTC explains ~12% of ZEC |
| Fisher 95% CI | [−0.11, +0.67] | [+0.83, +0.97] | ZEC interval spans 0 |
| Bootstrap 95% CI | [−0.05, +0.70] | — | consistent with 0 |
| Same-direction days | 13 / 21 (62%) | 20 / 21 (95%) | majority co-move |
Figure 3. September daily log returns. Left: ZEC versus BTC — a shallow positive fit (r = +0.34) with the eight opposite-sign days marked. Right: the ETH control, tightly aligned along the market factor.
Regressing ZEC returns on BTC returns gives a positive beta, not a negative one, but with a very wide interval:
| regression | alpha (daily) | beta | beta 95% CI | R² |
|---|---|---|---|---|
| ZEC ~ BTC, September | +2.60% (p=0.09) | +1.20 (p=0.16) | [−0.48, +2.88] | 0.12 |
| ZEC ~ BTC, full history | +0.55% (p=0.04) | +1.35 (p<0.001) | [+1.07, +1.63] | 0.20 |
| ETH ~ BTC, full history | −0.01% (n.s.) | +1.36 (p<0.001) | [+1.25, +1.46] | 0.72 |
The positive alpha is the elephant: ZEC delivered returns unrelated to BTC. Over the full record that alpha is ~0.55%/day with p=0.04 and a beta above one — high volatility, high idiosyncratic drift, low explanatory power.
Eight of twenty-one sessions moved in opposite directions. They cluster late in the month and are the likely source of the “opposite correlation” impression.
A trailing 30-day correlation over the full record never turns negative. It ranges from +0.06 (30 Nov 2025) to +0.84 (4 Feb 2025), averaging +0.51. September’s trailing average is +0.58 — on the high side of the asset’s own history, not the low.
Figure 4. Rolling 30-day return correlation, BTC–ZEC (red) and BTC–ETH control (blue). The zero line is never crossed by either pair. Shaded region: September 2026.
Across 21 calendar months the BTC–ZEC return correlation is never negative. The weakest month is November 2025 (+0.06); September 2026 (+0.36) ranks in the lower third but remains positive.
ZEC vs BTChighlighted = Sep 2026ETH vs BTC (control)
Figure 5. Monthly return correlation. ZEC (red bars) versus the ETH market-factor control (blue dots). September 2026 is highlighted. No negative month exists.
Cross-correlation is essentially zero at every non-zero lag: the relationship is purely contemporaneous, so neither asset predicts the other.
Figure 6. corr(ZECt, BTCt+k) by lag k. Only k=0 is material (+0.44, full sample); all |k|≥1 are within ±0.05. September (hollow) is noise-dominated at n≤21.
Conditioning on BTC’s direction does not rescue the hypothesis. Pooled over the full history, ZEC’s correlation with BTC is +0.26 on BTC-up days and +0.25 on BTC-down days — symmetric and positive. In September the tiny sub-samples split to −0.01 (11 up days) and +0.45 (10 down days), a difference produced by sampling noise, not by an inversion.
ZEC’s most extreme sessions are idiosyncratic: on 1 Oct 2025 it rose +62.9% while BTC added +3.9%; on 8 Oct 2025 it added +39.5% against +1.6% for BTC. These are ZEC-specific repricings, and they pull the full-sample correlation down, not up.
Leave-one-out on September confirms how fragile a 21-point estimate is: dropping 3 Sep raises the correlation to +0.52, while dropping 21 Sep lowers it to +0.24. The full-history estimate is far steadier (+0.43 to +0.46).
| date | ZEC return | BTC return | ETH return |
|---|---|---|---|
| 2025-10-01 | +62.9% | +3.9% | +4.8% |
| 2025-10-08 | +39.5% | +1.6% | +1.7% |
| 2026-06-04 | −26.0% | −0.4% | −2.4% |
| 2026-08-21 | +29.1% | +7.0% | +7.8% |
| 2025-10-11 | +29.0% | −1.9% | −2.0% |
| 2026-04-07 | +27.4% | +4.4% | +6.1% |
The ETH–BTC control behaves exactly as a market factor should (+0.93 in September, +0.85 full). The method is sound; the market simply does not contain a BTC–ZEC inversion.
The suspected strong opposite correlation between BTCUSDT and ZECUSDT is not supported.
Do not trade a BTC–ZEC inverse thesis. If the goal is a hedge or pairs relationship, work with returns (never levels) over a longer sample, size for ZEC’s ~3.5× volatility, and treat its alpha as an idiosyncratic trend that may reverse. A proper next step is a rolling, HAC-corrected beta/hedge-ratio study on at least a year of hourly data.
# history (monthly archives)
quantlab data build binance --market futures --symbol BTCUSDT --interval 1d \
--start 2025-01-01 --end 2026-08-31 --granularity monthly
quantlab data build binance --market futures --symbol ZECUSDT --interval 1d \
--start 2025-01-01 --end 2026-08-31 --granularity monthly
quantlab data build binance --market futures --symbol ETHUSDT --interval 1d \
--start 2025-01-01 --end 2026-08-31 --granularity monthly
# September 2026 (daily archives)
quantlab data build binance --market futures --symbol BTCUSDT --interval 1d \
--start 2026-09-01 --end 2026-09-23 --granularity daily
quantlab data build binance --market futures --symbol ZECUSDT --interval 1d \
--start 2026-09-01 --end 2026-09-23 --granularity daily
quantlab data build binance --market futures --symbol ETHUSDT --interval 1d \
--start 2026-09-01 --end 2026-09-23 --granularity daily
quantlab data validate --symbol ZECUSDT --dataset bars --interval 1duv run python scripts/correlation_report.py \
--start 2025-01-01 --end 2026-09-23 --sept-start 2026-09-01 --window 30 \
--out experiments/reports/btc_zec_correlation| path | contents |
|---|---|
experiments/reports/btc_zec_correlation/summary.json | all statistics, rolling series, monthly table |
experiments/reports/btc_zec_correlation/series.csv | aligned prices and log returns |
scripts/correlation_report.py | analysis source |
Pearson/Spearman/Kendall via SciPy; Fisher and bootstrap intervals computed directly; HAC OLS and ADF via statsmodels.