Predicting the Health of Retail in 2025
What the most recent same-store sales growth data suggests for next year
I've been studying historical retail sales trends in support of a 2025 goal-setting project for a client of mine.
I generally prefer to use same-store sales growth figures to judge the health of a retail business, since these capture aggregate customer purchasing behavior, independent of where new locations are being opened or shut down.
For this analysis, I wanted a broad market data set. I began by scraping quarterly public company SEC filings for the last 20 years. There is no regulation requiring companies to post same-store sales figures, and they aren't a GAAP standardized field, but enough companies do publish them for it to be possible to aggregate them into a macro-economic indicator.
This is a revenue-weighted chart of year-over-year same-store sales growth for the 100 largest publicly traded GICS sector Consumer Goods companies that report same-store sales.
Included in the chart are companies like Walmart, Starbucks, Nike, and Auto Max, among others. Because Walmart is large, it counts for more than the others. A given point on the line is a proxy for how much $ revenue growth there was in a representative retail store relative to the past year.
One of the first things that jumps out to me is that sales growth figures in the 2020-2021 period are the highest of the series. While I buy that there was a big purchasing spike coincident with the second round of stimulus checks in December 2020, I was a bit surprised to see how long into 2021 that spike lasted. There might be some data artifacts there, where many retailers chose not to disclose same-store sales growth for Q2 2020, but then did disclose in Q2 2021, showing huge double digit increases compared to those quarters impacted by lock down mandates.
I also see what looks like seasonal effects, especially in the first few years of the data series. Having any seasonal impact is a bit surprising since the data points are all year-over-year quarterly comparisons.
To get a clearer picture of industry trends, I made two adjustments. First, I normalized the data to CPI — this produces an inflation-adjusted retail sales growth series — and second, I smoothed the data across three quarters.
Cleaned up in this way, a few things become immediately apparent:
Nominal same-store sales has not contracted for retail as a whole except during the GFC in 2009
Contraction in real sales has happened briefly several times, but contraction of more than -2% has only occurred in the run-up to the GFC and over the last two years (2022-2024)
The latest 2024 same-store sales figures (published in November/December for quarters ending September/October) suggest an industry that is doing worse than in most periods of the last twenty years, but clearly still better than during the GFC
The real sales growth component is particularly important because it is a proxy for sales volume, which many retailers use as a KPI, either directly or in the form of a turn metric.
As a sanity check, I also looked at two other normalization factors. Here are the same figures also normalized to core PCE and M2 money supply
Core PCE tracks reasonably well with CPI. The M2 money supply chart diverges, especially in the pandemic period, but because changes in M2 most immediately impact large loans and transactions between institutions, it’s probably not relevant for consumers. The M2-adjusted series is effectively a proxy for “What percent of all liquidity in the dollar system is being used for retail purchases?” which may be interesting to follow up on, but isn’t very relevant here.
My client is an apparel and outdoor gear retailer, and while I can’t filter down to just apparel and specialty retail companies without constricting the sample size to a level where statistical tools are not useful, I can at least move one layer down the GICS classification scheme to the Consumer Discretionary industry group of companies. This smaller group excludes Consumer Staples companies like Costco.
This chart broadly tracks the full retail set, but for discretionary goods, we see a definite decline in both nominal and real same-store sales in the post-COVID period.
Because of the massive spike followed by contraction in 2021-2023, I’d be leery of calling for a recession based solely on the declining same-store sales. It looks to me like the contraction is driven in large part by an artificially high denominator during the stimulus boom. That said, the broad observation of “retail today looks less healthy than usual, but not as bad as the GFC,” appears to hold for discretionary as well as staple consumer goods.
On to forecasting 2025
When you’re trying to predict the next value in a time series, there are really two categories of approaches:
You can treat the series as a whole, de-emphasizing the immediately prior data points and instead fitting all historical data to a distribution, followed by randomly sampling to identify a plausible range of future values. This makes sense when the data is very noisy or when there’s not a causal reason to believe one data point will resemble its immediate predecessor
You can prioritize the most recent data points over the older ones, and then either fit a function to them to extrapolate forward, or use some form of price action technical analysis to generate a future expectation value
Technical analysis doesn’t make sense on a relative growth series (and many common forms of technical analysis used by retail investors are statistically unsound), but the chart does clearly show some data dependence on immediately prior points.
I think the best path forward is to use approach 2 in a way that decays into approach 1 (long-term statistical average) in the distant future. The question of “How far should I extrapolate trends into the future, before reverting to the mean?” is a fascinating topic, but my own data set is not large enough to warrant a rigorous approach, so I’m eyeballing it.
Judging from the the chart, it seems that 18 months is an upper bound on how far back preceding data might be relevant to predicting the next data point. Because of that, I’ll fit a model such that by 6/30/2026, the expectation value is simply the 20-year growth average.
I start by fitting both nominal and real sales to a Gaussian to get the longer-term expectation value and standard deviation.
Both series are moderately well described by normal distributions, with the nominal sales observations being a slightly better fit.
I think it’s worth noting that these Gaussian functions drastically underestimate the likelihood of very bad quarters (< -5% same-store sales growth). In practical terms, this means that my model will only ever predict a terrible quarter if you have several quarters of increasingly bad data beforehand.
Onto the second step, which is fitting the most recent observations to a function. It’s not immediately clear to me whether I would obtain more predictive power from taking the average of the last X quarters, or fitting a linear regression to the last X quarters. To find out, I run a set of tests on the full time series, capturing correlation coefficients across two variables (simple average vs. linear, and number of quarters of lookback) for both the nominal and real same-store sales series.
The 2-quarter views seem to be generally worse than the 4-quarter views, and the simple average and linear fits are about equal for 4+ quarters of lookback. As a compromise, I chose a model based on the 4-quarter lookback and came up with an algorithm for predicting the expectation value of each series.
For each future quarter:
Fit a linear regression to the previous 4 quarters and evolve it forward one time step
Take the average of the last 4 quarters
Average the resulting values in steps 1 and 2
For Q4 2024 (data to be published January-March 2025), use 100% weight on the value from step 3 and 0% weight on the long-term average produced by the Gaussian function fits (+2.2% nominal growth and -0.4% real growth), and take a weighted average
For each successive quarter, shift the weights such that by 6/30/2026, the weight is 100% on the long-term average
With that approach for the expectation value, I plot a chart, taking the standard deviation of the original data series to shade a ~2/3 confidence interval around the prediction
These two charts begin with the (non-smoothed) data points for 2022 through Q3 2024, and then forecast the next 7 quarters. Averaging the predictive model across 2025 yields an expectation value of +1.6% nominal and -1.9% real same-store sales growth for 2025.
Thoughts on goal-setting and managing to goals
My industry forecast is an external-factor-agnostic model. I didn’t build in any assumptions related to the impact of Fed policy or consumer revolving credit balances or future government stimulus or loan moratoria. This is simply a prediction of the next few points in series, given 20 years of history.
I opened this piece with the premise of using broad retail benchmarks and historical trends to help inform goal-setting for a particular retail company in calendar 2025.
For my particular use case, I incorporated both my predictions for the industry and historical data for their company to come up with targets for same-store sales growth, as well as turnover targets for regional and store managers. I recommended a set of targets that were around 2% higher (absolute comparison of nominal same-store sales growth) than the forecasted industry growth rates.
There’s a much larger conversation topic here related to setting and managing to goals. You might want conceptually to reward out-performance relative to the industry, and so be led towards designing a bonus plan that pays out relative to these benchmarks. The problem with that is that as a manager, you can’t course correct to achieve relative performance. It’s both a lagging indicator and a moving target. KPIs under you control that you can plausibly manage to are a particular store-wide revenue or gross margin target, or a particular turnover number.
A lot of my career has been spent working through the questions and tradeoffs that arise when setting and managing to quantitative goals. Writing December 30, 2024, we face a current economic moment marked by decidedly mixed signals. Namely:
Real GDP is growing moderately
The full-time employed population is contracting
S&P 500 earnings showed double digit growth in 2024, and the S&P 500 index is coming off its second straight year of >20% returns
Manufacturing PMIs are persistently below 50, indicating contraction
Existing home sales volume is at a 14-year low
The Fed recently began an easing cycle with a big 50 bps cut
I think the most likely place we are at in the business cycle is “late stage”, on a downtrend (either decreasing growth or outright contraction) that hasn’t yet turned over to the next expansion cycle. I’d place 50/50 odds that at least one quarter of 2025 will be backdated to be labeled as a recession. Note that my 50/50 odds prediction isn’t a cop-out. The long-term trend is for recessions to occur about 1/8 of the time, so my prediction is that 2025 is actually a 4x normal recession risk year.
Because of that, I think it’s more worthwhile than average to prepare for contingency scenarios.
I also recognize that with the White House and both chambers of Congress under the control of a single party, the likelihood for wild recession-averting spending increases is quite a bit higher than in years of divided government. A second implementation of Trump stimulus checks, for instance, would drastically shift the retail landscape overnight.
I ended up recommending to my client that their incentive cash programs be tied to absolute goals with a fallback condition that is triggered if both 1) industry benchmarks show negative nominal same-store sales in 2025 and 2) the company’s stores outperform the industry revenue-weighted average by at least 150 basis points. In this contingency, store managers, executives, and hourly retail staff will still be eligible to earn bonuses, even if they fall well short of their 2025 goals.
If you’re interested in talking through an issue like this for your own business, feel free to reach out. I offer free consultations for work in the goal-setting, macroeconomic impacts on businesses, and compensation incentive design space.
Key takeaways:
Same-store sales trends start from a worse-than-average place in 2024, but appear to be pointing towards moderation in 2025
I’m forecasting +1.6% nominal same-store sales growth for the revenue-weighted average of the Consumer Discretionary retail industry group. This is below long-term trend growth
Because there are a larger-than-average number of recession/contraction indicators flashing, I encourage retailers to have contingencies in place for if the economy dips into recession in 2025