Research paper · HKU

Designing Ethical Trading Apps

Mixed methods 6 min read

How UX design influences financial decision-making. I designed bias-exploiting and neutral versions of five trading screens, A/B tested them with 58 people, and turned the results into ethical guidelines for designers.

Read the full paper ↗
Bias-exploiting
Neutral
RoleResearcher & Designer
WhenSpring 2025
MethodsCase studies, A/B survey,
t-tests, regression
ToolsFigma, Qualtrics, Excel
context Made for DESN2003 Research for Innovation at the University of Hong Kong Final finding report · BASc Design+ · Sem 2, 2024–25.
with gratitude Thank you, Dr. Hongshan Guo My professor and supervisor, for the guidance, thoughtful feedback and encouragement that shaped this research from first question to final analysis.
the problem

Trading apps use the same persuasion tricks as social media, with much less scrutiny

Zero commission, one-tap trading and confetti animations made markets accessible. They also push people towards impulsive, excessive trading. Designers already work with cognitive biases, but there are few ethical rules for doing it where real money is at stake.

Research question

Which interface elements amplify cognitive biases in trading apps, and how can designers use bias-aware design without taking away users’ autonomy?

dark patterns in the wild

Users spot them on the apps they use every day

The literature names three big categories: hidden costs, forced continuity and gamification. People have already called out examples on mainstream platforms.

YouTube dark pattern screenshot
YouTube Buttons for different actions look the same, which makes choosing harder.
Amazon dark pattern screenshot
Amazon Look-alike options blur which one you’re actually choosing.
LinkedIn forced continuity tweet
LinkedIn Forced continuity: turning off email notifications takes 64 menus.
Instagram Shop tab screenshot
Instagram Put Shop where Likes used to be, relying on users’ muscle memory.
case study analysis

Three platforms, from persuasive to protective

Persuasive Protective
Gamified · retail Robinhood Zero commission, one-tap trades and instant execution make trading frictionless. Reward animations, bright colours and real-time alerts can push people to overtrade.
Global crypto exchange Binance In between the two. It mixes engagement-driven patterns with more protective ones.
Long-term investing Vanguard A calmer interface built around long-term investing, with more user-protective strategies.

I looked at pop-ups and notifications (their content and tone) and how saturated colours were used. Then I coded each platform on three ethical dimensions: transparency, autonomy and risk mitigation.

method

From literature to a controlled experiment

1 Literature review Defined healthy and unhealthy trading habits. Picked five biases with the biggest effect on financial UX: anchoring, loss aversion, overconfidence, recency and choice overload.
2 Case study analysis Compared Robinhood, Binance and Vanguard to find the UI elements that trigger unhealthy trading.
3 Feature testing survey A two-minute Qualtrics survey. A randomiser split people into a control group (neutral screens) and a treatment group (bias-exploiting screens). After each screen they rated how likely they were to invest. The screens were original, not copies of any real platform.
4 Reflection & analysis Post-survey questions asked what had influenced people and whether they spotted any biases. Then came descriptive stats, independent t-tests with Cohen’s d, and a logistic regression.
the experiment

Five biases, two versions of each screen

Both versions are identical except for one element, the variable being tested. I redesigned them here to current iOS conventions.

{{ cur.name }}
What it is{{ cur.def }}
How I triggered it{{ cur.sim }}
{{ cur.sigLabel }} {{ cur.stat }}
Bias-exploiting
Neutral
who took part
58participants, randomly split
79%aged 18–24
78%beginner traders
34%described themselves as somewhat risk-averse
results

Two of the five biases moved behaviour significantly

Mean investment likelihood (1–5)
Bias UINeutral UI
{{ r.b }}
{{ r.n }}
{{ r.short }}
Key takeaway In this sample, a single UI change was linked to a medium-sized shift in two of five tests. A celebratory pop-up was associated with higher willingness to invest (4.0 vs 3.5), while a crowded coin grid with lower willingness (3.0 vs 3.4). The other three showed no clear effect.
Means converted from Likert responses (1 = very unlikely, 5 = very likely). Values are approximate.
Independent t-tests
Biastpdfd
{{ r.name }}{{ r.t }}{{ r.p }}{{ r.df }}{{ r.d }} {{ r.sigShort }}
Key takeaway Not every bias is equally exploitable. Only 2 of 5 held up statistically. The ones that did act on feelings and effort rather than on numbers.

Overconfidence raised investment likelihood. Choice overload lowered it. Both were medium effects (d ≈ 0.55). Anchoring, loss aversion and recency showed no significant difference.

Logistic regression · overconfidence The interface predicted behaviour, and risk-tolerant users were more susceptible
UI type (bias / neutral)β 1.58p .004Significant
Risk toleranceβ 0.93p .022Significant
Trading frequencyβ 0.37p .134n.s.
McFadden’s R² = 0.913
32% spotted a bias in the interface People most likely to invest after the overconfidence screen mostly said they saw no bias at all.
Fear of loss was the factor people said influenced them most, followed by recent price surges. Clear performance data was what would make them invest.
the surprise The neutral screen sometimes persuaded more

In the anchoring test, about 12% more people chose “Likely” on the screen with no bias. Simple, pressure-free interfaces may come across as more trustworthy, which means even “neutral” design shapes decisions.

the framework

Six guidelines for designing trading platforms ethically

{{ g.n }} {{ g.title }} {{ g.strategy }} e.g. {{ g.example }}
{{ t }}
limitations & next steps
Small sample58 participants limits statistical power, especially for detecting smaller effects.
Mostly beginnersNext time I’d use stratified sampling across beginner, intermediate and professional traders.
Five biases, three platformsOther biases and newer platforms with different strategies are worth testing.
the full paper This page is the short version. The paper shows how I got there. 29 pages on why some biases moved people and others didn't, and what that means if you design anything involving money.
01The full literature review on dark patterns and biases 02The Robinhood, Binance and Vanguard breakdown 03Every survey screen and response breakdown 04Complete statistics and references