Summit White Paper Series: Adobe Case Study

When Estimating Reported Revenue Prior to its Release, Do Summit Data Have Statistically Significant Value?

Stu Jenkins | January 31, 2026

Research Question

Earnings surprise is a primary driver for short-term equity price reaction. Alternative data products aimed at predicting this error term are dominated by consumer credit card panels; B2B panels receive comparatively less systematic evaluation as far fewer firms have access to reliable data sources. This paper asks: Do Summit Data Have Statistically Significant Value? Value is defined as a non-zero, statistically significant Pearson correlation between multiple EMA-derived Summit features and the consensus error revealed after fiscal quarter-end.

Data and Sample Construction

Credit card panel. Daily Adobe transactions span January 2019 — January 2026. A recency-frequency cadence filter requires at least one distinct purchase per six-month period across a three-year lookback, yielding a cohort of loyal buyers with daily total spend and per-user rate metrics.

Summit (B2B invoice panel). Summit records episodic invoice loads submitted by Adobe to a large corporate counterparty. Net spend is $\text{net\_usd} = \text{gross\_usd} + \text{adjustment\_usd}$ (credits are negative). Days without a load receive $0 before smoothing: absence is a genuine zero, not missing data.

Earnings error term. The dependent variable is the consensus error:

\varepsilon = R_{\text{Adobe}} - C_{\text{VA}}

where $R_{\text{Adobe}}$ is the revenue reported by Adobe during each earnings call and $C_{\text{VA}}$ is the Visible Alpha market consensus prior to the earnings call. We evaluate both the signed error $\varepsilon$ and its absolute value $|\varepsilon|$ to separately identify directional correctness and error magnitude, a proxy for expected volatility.

Signal Construction

We apply five standard EMAs (spans: 7, 30, 91, 182, 365 days) to the null-filled Summit net USD series using the standard recursive form:

\text{EMA}_t = \alpha \, x_t + (1 - \alpha) \cdot \text{EMA}_{t-1}, \quad \alpha = \frac{2}{s + 1}

and construct all ten pairwise crossover spreads $\text{EMA}_{\text{fast}} - \text{EMA}_{\text{slow}}$ .

Raw panel data for Summit and credit card series

Results

Signal	$r$ (signed $\varepsilon$ )	p-value	sig	$r$ ( $\lvert\varepsilon\rvert$ )	p-value	sig
EMA-182d	0.464	0.011	*	0.454	0.013	*
EMA-91d	0.457	0.013	*	0.453	0.013	*
EMA-182d vs. 365d	0.431	0.020	*	0.450	0.015	*
EMA-365d	0.400	0.031	*	0.379	0.042	*
EMA-7d	—	n.s.		0.370	0.047	*
EMA-30d	—	n.s.		—	n.s.

Medium-to-long-horizon EMA levels (91d, 182d, 365d) are the strongest predictors of the consensus error. The 182-day EMA achieves the largest $r$ in both specifications ( $r = 0.46$ signed, $r = 0.45$ absolute), consistent with a two-quarter smoothing window capturing the persistent component of B2B activity. Thirty-day signals are uniformly insignificant, indicating that monthly B2B fluctuations are noise rather than demand signal.

Summit EMA levels overlaid with earnings surprise markers

Conclusion

Summit data are a statistically significant value-add for predicting consensus error. Medium-to-long-horizon EMA levels of Adobe’s B2B invoice volume yield Pearson $r \approx 0.40$ — $0.46$ ( $p < 0.05$ ) against both signed and absolute earnings surprise, despite Adobe’s revenues being dominated by subscription contracts rather than discrete invoicing. Future work should condition on myriad credit card panel features and assess stability across Adobe’s fiscal regimes.

References

American Psychological Association (2020). Publication Manual (7th ed.). APA.
Appel, G. (2005). Technical Analysis: Power Tools for Active Investors. FT Prentice Hall.
Fader, P. S., Hardie, B. G. S., & Lee, K. L. (2005). Counting your customers the easy way. Marketing Science, 24(2), 275—284.
Hughes, A. M. (1994). Strategic Database Marketing. Probus.
Pearson, K. (1895). Proceedings of the Royal Society of London, 58, 240—242.