Predictability over Volatility: When simple strategies help households save on fuel
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.
Tom: I'm Tom.
Jane: And I'm Jane, here with Lu, Meng and our in-house large language model, Lalam.
Tom: We're looking at the paper: "Predictability over Volatility".
Jane: Here's the gist of the paper. What reasonable strategies can households use to save on gasoline or diesel purchases under different pricing functions?
Tom: We assess basic strategies households might employ under a simplified model of a 10-gallon tank, fixed fuel consumption between refill opportunities, and price changes at each opportunity to refill.
Jane: We explore three strategy types: price-independent (filling once the tank is below a specific level), price-dependent (filling when the price is below the last n days or their average), and mixed (filling only below a set level, and then only when the price is below the last n days or their average).
Tom: We find that price-independent strategies are effective for the spline and linear functions, which have low volatility, while price-dependent strategies are effective for the normal and uniform distributions, which are highly volatile.
Title and authors: Tom: So we’re diving into the paper titled "Predictability over Volatility: When simple strategies help households save on fuel." We already touched on how they test different buying rules against various price models, but this section is about who wrote this and what it's really trying to get at.
Jane: The authors are Shahrzad Jamshidi and Ako Ogihara, from Mathematics and Computer Science at Lake Forest College. They’re applying these math concepts to a very real-world problem: helping people save on fuel when they can’t afford complex forecasting models.
Lu: The paper is trying to show that households don't need fancy computer programs to get a small edge on fuel costs, but they need the right rule for the market conditions. It’s about using simple heuristics that work best under specific price dynamics, which is what they call predictability over volatility.
Meng: If you think about it from an engineering standpoint, this isn't just a theoretical exercise; it's about creating something usable that works even when the data is messy and we don't have perfect forecasts. The authors are trying to make simple rules robust enough for real-world use.
Lalam: I see how this matters because it moves the focus away from trying to predict every single price point and toward understanding the structure of price movement itself, which is a much more manageable problem for a household strategy.
Tom: Right, so we're setting up the context here. The paper isn't just throwing numbers at you; it’s framing this as a decision-making problem under uncertainty where predictability matters more than just how shaky the price movement is. Where does that lead us?
The paper's summary: Jane: To summarize what they did, the authors constructed price functions from real US weekly averages for regular gasoline and diesel. They then simulated these prices using three distinct mathematical models: a smoothing spline, linear interpolation, and two different random draws from normal and uniform distributions.
Tom: So we have these scenarios where prices are very stable but trending up with the spline model, versus scenarios where they are super volatile with the Brownian Motion or the random draws. The goal was to see which buying strategies—price-independent, price-dependent, or mixed—did best in each of those specific market conditions.
Lu: The key finding here is that for the stable and trending models like the spline and linear interpolation, price-independent strategies tend to perform strongest. They are effective when the prices are predictable because they don't need a complex history to make a call.
Meng: But then they switch gears completely for those highly volatile models, like the normal and uniform distributions. In those cases, price-dependent strategies become significantly better at saving money compared to just filling up when you hit a fixed level.
Lalam: And this is the big point: the paper finds that price-dependent strategies are strongest precisely where volatility is highest, meaning you need more information about recent prices to make a good choice when things are jumping around a lot.
Tom: So, what they're saying simply is that your best strategy depends entirely on whether prices are trending or just random noise, not just how much the price is swinging up and down. This really changes how we think about managing fuel costs.
The paper's improvements: Jane: So, what do the authors suggest we should take away from this research? They point out that their main improvement is showing that you need to distinguish between predictability and volatility when choosing a rule. It’s not just about which price model looks best in isolation.
Tom: They suggest moving toward adaptive purchasing heuristics. Instead of using one single rule for everything, households should match the strategy type to the price process characteristics they are seeing in the market right now.
Lu: The authors propose a better way to approach this by recognizing that different price processes—like Brownian Motion versus a smoothing spline—are not just different levels of volatility on a single scale. They are fundamentally different types of stochastic processes, and the model needs to reflect that difference.
Meng: From an operational standpoint, this means if you’re designing an app or a tool for budgeting, you shouldn't use one fixed rule for everyone; it needs to dynamically switch its logic based on whether the data looks like a trend or noise.
Lalam: And this leads to the idea that policymakers should look at how prices evolve from one period to the next, not just what the price level is at any given moment. That’s a big shift in focus for them.
Tom: So, they are pushing for a system that can detect which regime you're in and automatically select the most appropriate simple strategy. That sounds like a much smarter way to approach saving fuel costs than just picking one fixed rule from the start.
Conclusion: Jane: So to wrap up, the paper "Predictability over Volatility: When simple strategies help households save on fuel" shows that your best savings strategy depends entirely on whether prices are trending or just random noise. It’s about matching your buying rule to the underlying price movement structure.
Tom: We learned that price-independent strategies work well when prices are stable and trending, while price-dependent ones shine when the market is very volatile and fluctuating a lot. The paper proves that predictability matters more than volatility for household decisions.
Lu: And they emphasize that we should look at how prices move from one period to the next instead of just looking at where the price is today. This moves us toward having adaptive systems that change their strategy based on whether the market is trending or reverting to a stable level.
Meng: I agree, and for deployment, this means building in that regime detection logic so the system knows when it needs to switch from a trend-based rule to a volatility-based one. Otherwise, you’re just applying the wrong heuristic.
Lalam: And I'm pretty sure this is where we should be focusing our attention with these kinds of papers today. It shifts the conversation from just setting retail prices to how those prices actually evolve in time, which is what really matters for household savings.
Shahrzad Jamshidi, *, Ako Ogihara
Lake Forest College
stat.AP
Submitted: 2026-11-03
Updated: 2026-11-03
License: http://creativecommons.org/licenses/by-nc-nd/4.0/
Importance score: 78/100
The gist: What reasonable strategies can households use to save on gasoline or diesel purchases under different pricing functions? We assess basic strategies households might employ under a simplified model of…
Key concepts
- Price-Independent Strategy
- This strategy involves filling up when the fuel level drops below a fixed threshold, regardless of the current price. It is most effective when prices follow smooth, predictable patterns like linear or spline functions where volatility is low.
- Price-Dependent Strategy
- This strategy dictates buying fuel only when the price falls below a certain benchmark, such as the average price over the last few days or a specific recent period. These are most useful when prices follow highly volatile distributions.
- Predictability vs. Volatility
- The research shows that whether prices trend predictably (move consistently in one direction) or revert to a stable level is more important for households than how wildly the prices fluctuate overall. Random walks, which are volatile but unpredictable, offer no usable signal for timing purchases.
- Price Process Regimes
- Prices are modeled using different mathematical processes, such as smoothing splines (for trends), normal/uniform distributions (for volatility), and Brownian motion (random walk). The paper distinguishes between these distinct processes to determine which saving strategy is appropriate for the specific market behavior.
Terminology
Summary
What reasonable strategies can households use to save on gasoline or diesel purchases under different pricing functions? p. 0 Predictability over Volatility: When Simple
Strategies Help Households Save on Fuel
Shahrzad Jamshidi1
,
*
Ako Ogihara2
1 Mathematics and Computer Science, Lake Forest College
2 Department of Economics, Business, and Finance, Lake Forest College
Abstract
What reasonable strategies can households use to save on gasoline or diesel purchases under different pricing functions? We assess basic strategies households might employ under a simplified model of a 10-gallon tank, fixed fuel consumption
between refill opportunities, and price changes at each opportunity to refill. Predictability over Volatility: When Simple
Strategies Help Households Save on Fuel
Shahrzad Jamshidi1
,
*
Ako Ogihara2
1 Mathematics and Computer Science, Lake Forest College
2 Department of Economics, Business, and Finance, Lake Forest College
Abstract
What reasonable strategies can households use to save on gasoline or diesel purchases under different pricing functions? We assess basic strategies households might employ under a simplified model of a 10-gallon tank, fixed fuel consumption
between refill opportunities, and price changes at each opportunity to refill. Purchases always fill the tank, and if the tank
is empty, it must be refilled regardless of price. We explore three strategy types: price-independent (filling once the tank
is below a specific level), price-dependent (filling when the price is below the last n days or their average), and mixed
(filling only below a set level, and then only when the price is below the last n days or their average). Together, these pricing models span a wide range
of volatilities. We find that price-independent strategies are effective for the spline and linear functions, which have low
volatility, while price-dependent strategies are effective for the normal and uniform distributions, which are highly volatile. Brownian motion, despite its mid-range volatility, admits no effective strategy. Our findings suggest that the predictability
of price movements—whether prices trend or revert toward a stable level—matters more for household strategies than overall
volatility: the random walk is volatile in the middle of our range yet offers households no usable signal. Our findings suggest that the predictability
of price movements—whether prices trend or revert toward a stable level—matters more for household strategies than overall
volatility: the random walk is volatile in the middle of our range yet offers households no usable signal. Policymakers can
therefore assist households by addressing how prices move from day to day, not only the retail prices they face.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Modeling Price Dynamics
The study constructs price functions from the nearest-in-time US weekly averages of regular gasoline and diesel, filling in higher-frequency values with a smoothing spline, linear interpolation, and random draws from normal and uniform distributions based on national averages. Additional simulations are built using Brownian motion. p. 0 We construct price
functions from the nearest-in-time US weekly averages of regular gasoline and diesel, filling in higher-frequency values
with a smoothing spline, linear interpolation, and random draws from normal and uniform distributions based on national
averages. Additional simulations are built using Brownian motion. Additional simulations are built using Brownian motion. Together, these pricing models span a wide range
of volatilities. We construct price
functions from the nearest-in-time US weekly averages of regular gasoline and diesel, filling in higher-frequency values
with a smoothing spline, linear interpolation, and random draws from normal and uniform distributions based on national
averages. Additional simulations are built using Brownian motion. 0251 Uniform Sampling (3 wks)
0. 0377 Uniform Sampling (5 wks)
Table 1: Generated datasets ranked by the statistical volatility,
which we calculate to be the (non-annualized) standard deviation
of the percent change, using a daily timescale. Each has 729 points, arising from
the 105 samples present in [3]. The mean absolute weekly changes in prices within
the original data are observed to be $0. The mean absolute weekly changes in prices within
the original data are observed to be $0. 3850%)
and $0. 0504%) for Regular (REG) and Diesel
(DSL), respectively. The non-annualized statistical
volatilities, measured as the standard deviation of
the weekly percent change, are 0. Weekly Statistical volatility observed
from [3] are 0. 0172 (REG) and 0. 0172 (REG) and 0. 0135 (DSL).Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Strategy Efficacy Across Regimes
The results are best broken up into three groups based on the datasets: curve fitting, distributional, and Brownian. No strategy was effective for all types because the Brownian datasets had no autocorrelation, meaning that historical prices offered no real information on future prices. p. 4 Simulations and Analysis
The results are best broken up into three groups
based on the datasets: curve fitting, distributional,
and Brownian. No strategy was effective for all types
because the Brownian datasets had no autocorrelation, meaning that historical prices offered no real
information on future prices. No strategy was effective for all types
because the Brownian datasets had no autocorrelation, meaning that historical prices offered no real
information on future prices. For the same reasons,
regular purchasing did not prove effective within the
random walk price model. The extreme values for these
functions that are differentiable almost everywhere
happen at the ends of the intervals. Given the downward trend in [3], favoring the end of the weeks will
generally be preferred.Where the paper says it
Where the paper says it
fill-at
strategies demonstrated the strongest performance, particularly those using levels of 33% and 10% with lags of 6 days or 5 days. Given the downward trend in [3], favoring the end of the weeks will generally be preferred. p. 4Where the paper says it
Exploiting Dispersion vs. p. 7 In the distributional datasets, consecutive prices are independent draws from a locally stationary distribution:
there is no momentum to ride, but there is a stable
marginal distribution with a well-defined mean and
spread. Here the exploitable quantity is not the trend
but the dispersion.Where the paper says it
In the distributional datasets, all strategies witnessed statistically significant results and all strategies saw a lower mean than the random purchasing distribution. The best performing strategies were the price-dependent strategies: Normal Distribution Draws and Uniform Distribution Draws. p. 4 The best performing strategies were the pricedependent strategies:
• Normal Distribution Draws: Strategies that purchased fuel when the price was lower than the
last n days performed best, where for all values
of n beat random purchasing. • Uniform Distribution Draws: Strategies that
purchased fuel when the price was lower than
the n-day average performed best, where all values of n beat random purchasing. with mean µ and standard deviation σ, the expected value of the window minimum
can be written
E
min
1≤i≤n
Pi
= µ − σ an, (1)
where an is a positive constant that increases with
the window length n. The expected saving relative
to a typical purchase is thus proportional to σ: the
wider the price distribution, the more a low-tail draw
is worth waiting for. The expected saving relative
to a typical purchase is thus proportional to σ: the
wider the price distribution, the more a low-tail draw
is worth waiting for. This is why price-dependent
strategies are strongest precisely where volatility is
highest, and why the advantage is larger for the highdispersion uniform samples than for the narrower
normal ones.Where the paper says it
Where the paper says it
Where the paper says it
The Role of Predictability
The Brownian price path has independent increments, which makes it a martingale: absent drift, the best forecast of any future price given the entire history is simply the current price, E[Pt+kFt] = Pt. The level is highly persistent—today is close to yesterday—so the path can superficially resemble the autocorrelated real series, but the increments carry no information, the process has no mean to revert to, and there is no fixed marginal distribution to define what counts as “low. p. 7 A Brownian price path
has independent increments, which makes it a martingale: absent drift, the best forecast of any future price given the entire history is simply the current price, E[Pt+k
| Ft
] = Pt
. The level is highly
persistent—today is close to yesterday—so the path
can superficially resemble the autocorrelated real series, but the increments carry no information, the
process has no mean to revert to, and there is no
fixed marginal distribution to define what counts as
“low. ” Both mechanisms that worked elsewhere fail
at once: there is no persistent trend for a need-based
rule to lean on, and no stationary spread for a threshold rule to harvest. Any timing rule measurable with
respect to past prices therefore has zero expected
edge over random timing, which is exactly what Table 4 reports.Where the paper says it
Where the paper says it
Reconciliation and Policy Focus
The empirical data thus sit far from the random-walk regime and close to the trending, serially dependent regime captured by the spline and linear models. A need-based heuristic—refill when the tank is low, with a mild preference for the end of the weekly price cycle—requires no price tracking, is consistent with the bounded-rationality justification of Section, and is exactly the strategy our analysis finds robust under realistic, trending, autocorrelated prices. p. 7 This is reassuring for
households: the price model that best matches reality is also the one in which the simplest rule works. A need-based heuristic—refill when the tank is low,
with a mild preference for the end of the weekly
price cycle—requires no price tracking, is consistent
with the bounded-rationality justification of Section ,
and is exactly the strategy our analysis finds robust
under realistic, trending, autocorrelated prices. This supports the paper’s central claim. Because gasoline savings propagate into broader consumption and inflation expectations ([12]; [1]), policymakers concerned with household welfare have
reason to attend not only to the retail price level but
8 Preprint Implications for Policy
The same regime distinction reframes the policy question. Because efficacy is governed by the temporal structure of prices rather than by their volatility,
interventions that change how prices evolve from
one period to the next can change which household
strategies pay off, independently of the price level.Where the paper says it
Where the paper says it
Where the paper says it
Limitations and Conclusion
The reference distribution is a mean of 1000 random-purchase trials and is approximately normal by the Central Limit Theorem, confirmed by the Shapiro–Wilk and Kolmogorov–Smirnov tests, so the “beats 95% of random” criterion functions as a one-sided test at α = 0.05 against that benchmark; it identifies strategies that are statistically, not necessarily economically, superior. p. 8 The reference distribution is a mean of 1000 random-purchase
trials and is approximately normal by the Central
Limit Theorem, confirmed by the Shapiro–Wilk and
Kolmogorov–Smirnov tests, so the “beats 95% of
random” criterion functions as a one-sided test at
α = 0. 05 against that benchmark; it identifies strategies that are statistically, not necessarily economically,
superior, and the realized dollar gaps implied by Table 3 are often small. 05 against that benchmark; it identifies strategies that are statistically, not necessarily economically,
superior, and the realized dollar gaps implied by Table 3 are often small. The model also abstracts from
station-to-station price dispersion, brand and quality
preferences, and the fixed time and travel costs of
more frequent stops, each of which would temper
the gains attributed to any active strategy. The model also abstracts from
station-to-station price dispersion, brand and quality
preferences, and the fixed time and travel costs of
more frequent stops, each of which would temper
the gains attributed to any active strategy. Finally,
the Brownian and distributional datasets are deliberately unrealistic instruments for isolating mechanism:
their value is diagnostic rather than predictive, and
it is the empirical comparison above—not the synthetic extremes—that anchors the recommendations
to observed prices. Holding this model and the space of purchasing
rules fixed, we vary the price process across a wide
range of volatilities, from smooth trends to pure random walks, and evaluate each rule against a randompurchasing benchmark. Our central finding is that
it is the predictability of price movements, not their
volatility, that determines whether a rule beats random buying: the random walk is volatile and so
offers no usable signal and so no strategy is successful.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
References
The references listed span literature on price asymmetry, behavioral models of rational choice, and empirical studies on gasoline retailing. The authors thank Nancy Tao, Dimitra Papadovasilaki, Tilahun Emiru and Linh Pham for valuable feedback on the project. p. 0 edu This work was
partially funded by the Richter Scholar Summer Research Program
at Lake Forest College. The authors thank Nancy Tao, Dimitra
Papadovasilaki, Tilahun Emiru and Linh Pham for valuable feedback on the project. Small reductions in the
*Corresponding author: sjamshidi@lakeforest. edu This work was
partially funded by the Richter Scholar Summer Research Program
at Lake Forest College. AI assistants were used in preparing this
manuscript: Le Chat (Mistral AI), Claude (Anthropic), and Apple
Intelligence (Apple) were used to help edit and refine drafts, and
Adobe’s generative AI tools were used in creating Figure 1. The
authors reviewed all AI-assisted output and take full responsibility
for the content. The response
of consumer spending to changes in gasoline
prices. Working Paper 22969, National Bureau
of Economic Research, December 2016.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improve decision-making under price processes by using a model that distinguishes between predictability and volatility, as
predictability of price movements—whether prices trend or revert toward a stable level—matters more for household strategies than overall volatility.
The improved system can assess which strategy works based on the underlying temporal structure of prices rather than just the magnitude of fluctuations. -
Develop adaptive purchasing heuristics by matching specific strategy types to price process characteristics, such as using
price-independent (filling once the tank is below a specific level)
for low-volatility trend models andprice-dependent (filling when the price is below the last n days or their average)
for highly volatile distributions. -
Enhance forecasting accuracy by applying econometric models to identify market regimes, specifically recognizing that
The spline, i.i.d. samples, and the random walk are not three points on a single volatility continuum; they are three distinct stochastic processes.
The system can then select the appropriate model (spline/linear for trends versus distributional models for stationary noise) to determine if a history-based rule offers anexploitable pattern.
-
Implement robust simulation capabilities by allowing users to test strategies across a wide range of price process volatilities, from
smooth trends to pure random walks,
while explicitly recognizing that under the Brownian price model,the Brownian group admits no effective strategy.
This prevents the system from incorrectly suggesting strategies based on volatility alone. -
Refine policy recommendations by focusing interventions on price evolution rather than just levels, as
interventions that change how prices evolve from one period to the next can change which household strategies pay off, independently of the price level.
The improved system can suggest policies aimed at fostering autocorrelation or mean reversion in markets to improve household savings.
Abstract
What reasonable strategies can households use to save on gasoline or diesel purchases under different pricing functions? We assess basic strategies households might employ under a simplified model of a 10-gallon tank, fixed fuel consumption between refill opportunities, and price changes at each opportunity to refill. Purchases always fill the tank, and if the tank is empty, it must be refilled regardless of price. We explore three strategy types: price-independent (filling once the tank is below a specific level), price-dependent (filling when the price is below the last n days or their average), and mixed (filling only below a set level, and then only when the price is below the last n days or their average). We construct price functions from the nearest-in-time US weekly averages of regular gasoline and diesel, filling in higher-frequency values with a smoothing spline, linear interpolation, and random draws from normal and uniform distributions based on national averages. Additional simulations are built using Brownian motion. Together, these pricing models span a wide range of volatilities. We find that price-independent strategies are effective for the spline and linear functions, which have low volatility, while price-dependent strategies are effective for the normal and uniform distributions, which are highly volatile. Brownian motion, despite its mid-range volatility, admits no effective strategy. Our findings suggest that the predictability of price movements, whether prices trend or revert toward a stable level, matters more for household strategies than overall volatility: the random walk is volatile in the middle of our range yet offers households no usable signal. Policymakers can therefore assist households by addressing how prices move from day to day, not only the retail prices they face.
Related papers
- Unraveling time-varying causal effects of multiple exposures: integrating Functional Data Analysis with Multivariable Mendelian Randomization
- Strategic Causal Policy Learning with Manipulable Eligibility
- BayesJudge: Uncertainty-Aware Bayesian Meta-Evaluation of Human and LLM Judgments
- Automating structural reliability analysis with a multi-agent large language model framework
- Improving Minority Population Sampling with BISG Probabilities: Evidence from a Survey of Jewish Americans
- When Do Sports Leagues Stop Being Competitive? Bayesian Detection of Competitive-Balance Eras from Match-Level Data