IELTS Reading: Regulating AI in Financial Markets – Đề Thi Mẫu Có Đáp Án Chi Tiết

Bài viết này cung cấp cho bạn một bộ đề thi IELTS Reading hoàn chỉnh với 3 passages theo đúng format thi thật, từ mức độ dễ đến khó, giúp bạn làm quen với các dạng câu hỏi đa dạng. Bạn sẽ nhận được đáp án chi tiết kèm giải thích cụ thể về cách xác định thông tin trong bài, các kỹ thuật paraphrase quan trọng, cùng với bộ từ vựng chuyên ngành thiết yếu. Đề thi này phù hợp cho học viên từ band 5.0 trở lên, giúp bạn rèn luyện kỹ năng đọc hiểu học thuật và quản lý thời gian hiệu quả trong bài thi thực tế.

Hướng Dẫn Làm Bài IELTS Reading

Tổng Quan Về IELTS Reading Test

IELTS Reading Test kéo dài 60 phút với 3 passages và tổng cộng 40 câu hỏi. Mỗi câu trả lời đúng được tính 1 điểm, không có điểm âm cho câu sai. Độ khó tăng dần từ Passage 1 đến Passage 3, yêu cầu bạn phải phân bổ thời gian hợp lý.

Phân bổ thời gian khuyến nghị:

• Passage 1: 15-17 phút (13 câu hỏi – mức độ dễ)
• Passage 2: 18-20 phút (13 câu hỏi – mức độ trung bình)
• Passage 3: 23-25 phút (14 câu hỏi – mức độ khó)

Lưu ý dành 2-3 phút cuối để chuyển đáp án vào answer sheet và kiểm tra lại những câu chưa chắc chắn.

Các Dạng Câu Hỏi Trong Đề Này

Đề thi mẫu này bao gồm 7 dạng câu hỏi phổ biến nhất trong IELTS Reading:

1. Multiple Choice – Câu hỏi trắc nghiệm
2. True/False/Not Given – Xác định thông tin đúng/sai/không có
3. Yes/No/Not Given – Xác định quan điểm tác giả
4. Matching Headings – Nối tiêu đề với đoạn văn
5. Sentence Completion – Hoàn thành câu
6. Matching Features – Nối thông tin với đặc điểm
7. Short-answer Questions – Câu hỏi trả lời ngắn

IELTS Reading Practice Test

PASSAGE 1 – The Rise of Algorithmic Trading

Độ khó: Easy (Band 5.0-6.5)

Thời gian đề xuất: 15-17 phút

The financial markets have undergone a dramatic transformation in recent decades, with artificial intelligence and computer algorithms playing an increasingly central role. Algorithmic trading, also known as automated trading or algo-trading, refers to the use of computer programs to execute trades based on predefined rules and strategies. This technology has revolutionized how financial markets operate, bringing both significant benefits and new challenges that regulators must address.

In the early 2000s, algorithmic trading accounted for less than 10% of total trading volume in major stock exchanges. Today, estimates suggest that automated systems are responsible for 60-75% of all equity trading in the United States and similar proportions in European markets. These algorithms can analyze vast amounts of data, identify trading opportunities, and execute orders in fractions of a second – far faster than any human trader could manage. The speed advantage is so significant that some trading firms invest millions of dollars in technology to reduce transaction times by mere microseconds.

High-frequency trading (HFT) represents the most sophisticated form of algorithmic trading. HFT firms use powerful computers and complex algorithms to make thousands or even millions of trades per day, holding positions for extremely short periods – sometimes just seconds or milliseconds. These firms often profit from tiny price differences that exist for very brief moments. For example, if a stock is priced slightly differently on two exchanges, an HFT algorithm can buy on the cheaper exchange and simultaneously sell on the more expensive one, capturing the price differential as profit. This practice, known as arbitrage, has existed for centuries, but modern technology has made it possible to exploit these opportunities at unprecedented speed and scale.

The benefits of algorithmic trading are substantial. First, it has significantly improved market liquidity – the ease with which assets can be bought and sold without causing major price changes. With more automated participants actively trading, there are more buyers and sellers in the market at any given time, making it easier for investors to execute trades quickly. Second, algorithms have reduced transaction costs for many market participants. The competition among algorithmic traders has narrowed bid-ask spreads (the difference between buying and selling prices), saving investors money on each trade. Third, automated systems eliminate emotional decision-making, executing trades based purely on data and predefined rules rather than fear or greed.

However, the rise of algorithmic trading has also introduced new risks to financial markets. One major concern is market volatility. While algorithms generally improve market stability, they can occasionally amplify problems during periods of stress. The most famous example is the “Flash Crash” of May 6, 2010, when the U.S. stock market experienced a sudden, severe drop. The Dow Jones Industrial Average plunged nearly 1,000 points (about 9% of its value) in just minutes before recovering most of the loss. Investigators determined that automated trading systems had exacerbated the decline, with algorithms triggering each other in a cascade of selling.

Another risk involves market manipulation. Some traders have used algorithms for illegal practices such as “spoofing” – placing large orders they intend to cancel to create false impressions about supply and demand, thereby manipulating prices. In 2015, British trader Navinder Sarao was arrested for allegedly using spoofing algorithms to contribute to the 2010 Flash Crash, earning millions of dollars in profits. Such cases highlight the need for effective regulatory oversight.

Fairness concerns also arise from algorithmic trading. Firms with the most advanced technology and fastest connections to exchanges gain significant advantages over traditional investors. This has led to debates about whether markets have become a “two-tier system” where those with sophisticated technology profit at the expense of ordinary investors. Some critics argue this undermines the fundamental principle of fair markets where all participants should have equal opportunity.

Given these challenges, financial regulators worldwide have begun developing frameworks to oversee algorithmic trading. Most major markets now require firms using trading algorithms to register with regulatory authorities and implement risk controls. These controls include automatic “kill switches” that can halt trading if systems malfunction and limits on the speed and volume of orders. Regulators also demand that firms test their algorithms thoroughly before deployment and maintain detailed records of algorithmic trading activity for supervisory review. Despite these efforts, regulating this rapidly evolving technology remains an ongoing challenge, requiring continuous adaptation as new trading strategies and technologies emerge.

Questions 1-6

Do the following statements agree with the information given in Passage 1?

Write:

• TRUE if the statement agrees with the information
• FALSE if the statement contradicts the information
• NOT GIVEN if there is no information on this

1. Algorithmic trading currently represents more than half of all equity trading in the United States.

2. High-frequency trading firms typically hold their positions for several hours before selling.

3. The Flash Crash of 2010 resulted in permanent losses for most investors.

4. Navinder Sarao was convicted of using spoofing algorithms.

5. Algorithmic trading has reduced the difference between buying and selling prices.

6. All countries have implemented identical regulations for algorithmic trading.

Questions 7-10

Complete the sentences below.

Choose NO MORE THAN TWO WORDS from the passage for each answer.

7. Some trading firms spend millions to reduce __ by microseconds.

8. The practice of profiting from brief price differences across exchanges is called __.

9. Automated trading systems remove __ from the trading process.

10. Regulators require firms to install automatic __ to stop trading during system failures.

Questions 11-13

Choose the correct letter, A, B, C, or D.

11. According to the passage, what is the main advantage of improved market liquidity?

    • A. It increases profits for algorithmic traders
    • B. It makes buying and selling assets easier and faster
    • C. It prevents market crashes
    • D. It eliminates the need for human traders

12. The “two-tier system” mentioned in the passage refers to:

    • A. Different types of algorithmic trading strategies
    • B. Separate markets for professional and retail investors
    • C. Advantages held by firms with superior technology
    • D. Two different regulatory frameworks

13. What does the passage suggest about regulating algorithmic trading?

    • A. Current regulations are completely effective
    • B. It remains a continuously evolving challenge
    • C. It is impossible to regulate effectively
    • D. Only the United States has proper regulations

---

PASSAGE 2 – Machine Learning and Risk Assessment

Độ khó: Medium (Band 6.0-7.5)

Thời gian đề xuất: 18-20 phút

Beyond trading, artificial intelligence is transforming another critical aspect of financial markets: risk assessment. Financial institutions have always needed to evaluate risk – whether assessing the creditworthiness of loan applicants, detecting potentially fraudulent transactions, or calculating the likelihood that investment portfolios will lose value. Traditionally, these assessments relied on statistical models developed by human analysts using historical data. Today, machine learning algorithms are increasingly handling these tasks, often with superior accuracy but also raising important questions about transparency, bias, and accountability.

Machine learning is a subset of AI that enables computers to learn from data without being explicitly programmed with rules. Instead of following predetermined instructions, these systems identify patterns in training data and use those patterns to make predictions about new data. In finance, machine learning models can process vastly more information than humans – analyzing not just traditional financial data but also alternative data sources such as social media activity, satellite imagery, and payment transaction patterns. This comprehensive analysis can reveal insights that human analysts might never discover.

Credit scoring provides a compelling example of machine learning’s potential. Traditional credit scores, such as the FICO score used widely in the United States, rely on a limited set of factors: payment history, amounts owed, length of credit history, types of credit, and new credit inquiries. These scores work reasonably well for individuals with substantial credit histories but often fail to accurately assess people with “thin files” – those with limited traditional credit data, including many young people and recent immigrants. Machine learning models can incorporate thousands of additional variables to assess creditworthiness, potentially extending credit access to populations previously excluded from traditional lending. For instance, algorithms might consider utility payment histories, educational background, or even patterns in mobile phone usage to predict loan repayment likelihood.

Several fintech companies have pioneered machine learning-based lending. Upstart, a U.S.-based lending platform, claims its AI models can approve 27% more borrowers than traditional methods while maintaining the same loss rates. Zest AI, another firm providing machine learning tools to lenders, reports that its models have helped financial institutions reduce loan losses by up to 25% while increasing approval rates. These improvements could democratize access to credit, particularly for underserved communities that traditional models often disadvantage.

However, machine learning in risk assessment presents significant regulatory challenges. The first concern involves algorithmic opacity. Many machine learning models, particularly those using deep learning techniques with neural networks, function as “black boxes.” Even their creators cannot fully explain how they arrive at specific decisions. When a traditional credit model denies a loan application, the lender can typically identify which factors led to the rejection. With complex machine learning systems, providing such explanations becomes extremely difficult. This opacity conflicts with regulatory requirements in many jurisdictions that give consumers rights to understand why they were denied credit.

The European Union’s General Data Protection Regulation (GDPR), implemented in 2018, includes a “right to explanation” for automated decisions. Financial institutions using machine learning in EU markets must somehow explain algorithmic decisions to customers – a requirement that has proven technically challenging for complex AI systems. U.S. regulations, while less explicit about algorithmic transparency, also require lenders to provide “adverse action notices” explaining credit denials. Reconciling these legal requirements with machine learning’s inherent complexity remains an ongoing struggle for both financial institutions and regulators.

A second major concern is algorithmic bias. Machine learning models learn from historical data, which may reflect past discrimination. If, for example, a bank’s historical lending data shows that fewer loans were granted to certain demographic groups due to discriminatory practices, a machine learning model trained on this data might perpetuate those biases. Even without explicitly considering protected characteristics like race or gender, algorithms can discriminate through proxy variables – other factors that correlate with protected characteristics. Zip code, for instance, might serve as a proxy for race in the United States due to residential segregation patterns.

Several real-world cases have demonstrated this risk. In 2019, the Apple Card, issued by Goldman Sachs and using algorithmic underwriting, faced allegations of gender bias after several users reported that women were offered lower credit limits than men with similar financial profiles. Regulators investigated these claims, highlighting the challenges of ensuring algorithmic fairness. Research has also shown that some facial recognition systems used for identity verification in financial services perform less accurately for people with darker skin tones, potentially creating discriminatory barriers to service access.

Addressing algorithmic bias requires multiple approaches. Diverse training data can help ensure models learn from more representative samples. Regular algorithmic audits – systematic testing to identify disparate impacts on different demographic groups – can catch problems before they harm consumers. Some researchers advocate for “fairness constraints” that prevent algorithms from producing outcomes with unacceptable disparities across groups, though implementing such constraints involves difficult trade-offs between fairness, accuracy, and different conceptions of what fairness means.

Accountability presents the third major regulatory challenge. When machine learning systems make consequential errors – denying credit to qualified applicants, failing to detect fraud, or miscalculating investment risks – who bears responsibility? Is it the financial institution deploying the algorithm, the software vendor who created it, the data providers whose information trained it, or the developers who designed it? Traditional liability frameworks struggle with these questions, particularly when multiple parties contribute to algorithmic systems and when the systems’ complexity makes it difficult to identify exactly what went wrong.

Regulators worldwide are developing responses to these challenges. The Monetary Authority of Singapore published “Principles to Promote Fairness, Ethics, Accountability and Transparency (FEAT) in the Use of AI and Data Analytics in Singapore’s Financial Sector” in 2018, providing guidance for financial institutions deploying AI. The principles emphasize proportionality – applying stricter oversight to higher-risk applications – and encourage firms to maintain human oversight of significant algorithmic decisions. The Bank of England and Financial Conduct Authority in the UK have similarly focused on ensuring firms understand and can explain their algorithmic systems, implement appropriate testing, and maintain human accountability for AI decisions.

These regulatory developments reflect an emerging consensus: machine learning offers tremendous benefits for financial risk assessment but requires careful oversight to ensure fairness, transparency, and accountability. Getting this balance right – enabling innovation while protecting consumers and maintaining market integrity – represents one of the most important challenges facing financial regulators today.

Questions 14-18

Choose the correct letter, A, B, C, or D.

14. According to the passage, machine learning differs from traditional statistical models primarily because:

    • A. It is more accurate in all situations
    • B. It can only process financial data
    • C. It identifies patterns without predetermined rules
    • D. It requires less training data

15. The passage mentions “thin files” to describe:

    • A. Physical documents used in traditional lending
    • B. People with limited traditional credit histories
    • C. Incomplete machine learning datasets
    • D. Simplified credit scoring algorithms

16. The “black box” problem in machine learning refers to:

    • A. The high cost of implementing AI systems
    • B. The difficulty of explaining how algorithms make decisions
    • C. Data storage requirements for neural networks
    • D. Security vulnerabilities in financial software

17. According to the passage, proxy variables in credit algorithms:

    • A. Always improve lending decisions
    • B. Are required by European regulations
    • C. Can indirectly discriminate against protected groups
    • D. Are the main cause of all algorithmic errors

18. The Singapore FEAT principles emphasize:

    • A. Banning all AI use in finance
    • B. Requiring identical regulations globally
    • C. Proportional oversight based on risk levels
    • D. Eliminating human involvement in decisions

Questions 19-23

Complete the summary below.

Choose NO MORE THAN TWO WORDS from the passage for each answer.

Machine learning has improved (19) __ in financial services, with companies like Upstart claiming to approve more borrowers while maintaining similar loss rates. However, these systems create regulatory challenges. The GDPR in Europe includes a (20) __ that requires companies to explain automated decisions. Another problem is (21) __, which occurs when models learn from historical data reflecting past discrimination. Several approaches can address this issue, including using more (22) __ and conducting regular (23) __ to identify disparate impacts.

Questions 24-26

Do the following statements agree with the claims of the writer in Passage 2?

Write:

• YES if the statement agrees with the claims of the writer
• NO if the statement contradicts the claims of the writer
• NOT GIVEN if it is impossible to say what the writer thinks about this

24. Machine learning models are always more accurate than traditional risk assessment methods.

25. The Apple Card investigation demonstrated real concerns about gender bias in algorithmic lending.

26. Most financial institutions have successfully resolved all transparency issues with machine learning.

---

PASSAGE 3 – Systemic Risk and Regulatory Frameworks

Độ khó: Hard (Band 7.0-9.0)

Thời gian đề xuất: 23-25 phút

The proliferation of artificial intelligence in financial markets extends far beyond individual transactions or isolated risk assessments; it fundamentally alters the systemic architecture of global finance, introducing novel forms of interconnected risk that transcend traditional regulatory paradigms. While previous sections have addressed AI’s impact on trading and credit decisions, the most profound regulatory challenge lies in understanding and mitigating how AI-driven systems might contribute to or exacerbate systemic crises – events that threaten the stability of entire financial systems rather than individual institutions. This dimension of AI regulation requires reconceptualizing how we understand financial stability in an era of algorithmic interdependence.

Systemic risk refers to the possibility that failures within individual components of a financial system could trigger cascading failures throughout the entire system. The 2008 financial crisis provided a stark illustration: the collapse of Lehman Brothers and troubles at other major institutions created a contagion effect that threatened the global financial system, requiring unprecedented government intervention. Regulators responded by designating certain institutions as “systemically important financial institutions (SIFIs)” subject to enhanced oversight. The underlying assumption was that size and interconnectedness – measured primarily through direct financial exposures between institutions – determined systemic importance.

AI introduces a fundamentally different type of systemic risk: technological homogeneity. When numerous financial institutions adopt similar AI models, train them on similar data, and employ them for similar purposes, they may inadvertently create correlated behavior across the financial system. This algorithmic monoculture could cause multiple institutions to make similar decisions simultaneously, potentially amplifying market movements and creating self-fulfilling prophecies. For instance, if many institutions deploy similar machine learning models for portfolio risk management, these systems might simultaneously recommend selling similar assets during market stress, accelerating price declines and potentially triggering a market collapse that might not have occurred with more diverse decision-making approaches.

Research by economists at the Bank for International Settlements (BIS) has highlighted this concern. In a 2019 paper, they note that “as AI systems become more prevalent in financial markets, there is a risk that herding behavior could increase, particularly during periods of stress when these systems might simultaneously identify similar risks and recommend similar responses.” The paper emphasizes that this risk differs from traditional systemic risk because institutions need not be directly connected through financial exposures to contribute to cascading failures; the mere similarity of their algorithmic systems creates indirect interconnections that traditional regulatory frameworks do not adequately capture.

Model risk – the possibility that models themselves contain errors or become obsolete – takes on systemic dimensions when combined with widespread AI adoption. Traditional financial models have occasionally failed spectacularly: the Black-Scholes option pricing model, while revolutionary, made assumptions that proved problematic during certain market conditions; the Value at Risk (VaR) models used by many banks before 2008 systematically underestimated the probability of extreme losses. These failures harmed individual institutions but occurred gradually enough that regulatory corrections could be implemented. With AI systems operating at high speed and scale, model failures could propagate more rapidly, potentially destabilizing markets before regulators can respond.

The opacity of advanced machine learning compounds this problem. Deep neural networks, which have proven remarkably effective for pattern recognition tasks, often contain millions or billions of parameters whose relationships are mathematically intractable to human analysis. Unlike traditional economic models based on explicit theoretical assumptions that domain experts can scrutinize and debate, these “black box” systems offer few opportunities for theoretical validation. An AI system might perform exceptionally well on historical data but fail catastrophically when confronted with unprecedented market conditions – precisely when accuracy matters most. The COVID-19 pandemic demonstrated this vulnerability: many algorithmic trading systems struggled when markets exhibited unusual patterns during March 2020, contributing to elevated volatility.

Cybersecurity represents another dimension of systemic AI risk. As financial institutions increasingly rely on AI systems for critical functions, these systems become attractive targets for malicious actors. A successful cyberattack that compromises widely used AI platforms could simultaneously disrupt numerous institutions. Moreover, the complexity of AI systems creates potential vulnerabilities that even their developers may not fully understand. Adversarial machine learning – techniques for deceiving AI systems through carefully crafted inputs – has demonstrated that algorithms can be manipulated in ways that are difficult to detect. In financial contexts, adversaries might exploit such vulnerabilities to manipulate trading algorithms, evade fraud detection systems, or disrupt risk management processes.

International regulatory coordination poses particular challenges for AI governance in financial markets. Financial services operate globally, with transactions flowing across borders instantaneously. Different regulatory jurisdictions, however, have adopted divergent approaches to AI oversight, creating regulatory arbitrage opportunities and compliance complexities for international financial institutions. The European Union has moved toward comprehensive AI regulation with its proposed Artificial Intelligence Act, which would classify AI systems by risk level and impose corresponding requirements. The United States, conversely, has favored a more sectoral approach, with different regulatory agencies addressing AI within their respective domains. China has implemented various AI regulations focused on algorithmic transparency and data governance, reflecting different priorities and political contexts.

These divergent regulatory philosophies reflect underlying tensions about AI governance that extend beyond finance. European approaches emphasize precautionary principles and individual rights protection, embodied in regulations like GDPR. American approaches traditionally favor innovation and market-based solutions, with regulation focused on addressing demonstrated harms. Chinese regulations balance innovation promotion with state oversight and social stability concerns. For financial institutions operating across these jurisdictions, regulatory fragmentation creates substantial challenges: AI systems must simultaneously comply with different and sometimes conflicting requirements, potentially limiting their effectiveness or forcing institutions to maintain separate systems for different markets.

Several international bodies are attempting to develop harmonized frameworks. The Financial Stability Board (FSB), which coordinates financial regulation among G20 countries, has made AI governance a priority. In 2017, the FSB established a framework for addressing third-party dependencies in financial services, recognizing that many institutions rely on a small number of technology providers for AI capabilities. This concentration creates systemic vulnerability: problems with major AI service providers could simultaneously affect numerous financial institutions. The International Organization of Securities Commissions (IOSCO) has similarly emphasized the need for cross-border cooperation in algorithmic trading oversight, noting that algorithmic systems operate across markets and jurisdictions in ways that challenge traditional regulatory boundaries.

Effective AI regulation in financial markets requires balancing competing objectives: promoting innovation while ensuring stability; protecting consumers while allowing data-driven insights; maintaining market efficiency while preventing manipulation. Some regulatory scholars advocate for “embedded regulation” – incorporating regulatory requirements directly into AI systems through technological means rather than relying solely on rules and oversight. For example, trading algorithms could contain built-in constraints preventing them from executing potentially manipulative strategies, or lending algorithms could include fairness checks that flag decisions with potentially discriminatory impacts. This approach, sometimes called “RegTech” (regulatory technology), could make compliance more efficient and effective.

However, embedded regulation raises its own concerns. Hardcoding regulatory requirements into algorithms might reduce flexibility needed to address novel situations or might ossify regulatory approaches, making them difficult to update as conditions change. There are also questions about who would design and validate these embedded controls, and whether the technological sophistication required might advantage large institutions over smaller ones, potentially reducing market competition.

Looking forward, regulatory frameworks for AI in financial markets will likely need to become more adaptive and technologically sophisticated. Traditional regulation operates on relatively slow cycles: rules are proposed, commented upon, finalized, and implemented over months or years. AI systems evolve much more rapidly, with models being updated continuously as they learn from new data. This temporal mismatch between regulatory and technological change suggests that future regulation might need to focus less on prescriptive rules and more on outcome-based standards, continuous monitoring, and adaptive processes that can evolve alongside the technology they govern. Some scholars propose “regulatory sandboxes” – controlled environments where financial institutions can test innovative AI applications under regulatory supervision before broader deployment, allowing regulators to understand new technologies and develop appropriate oversight frameworks.

The fundamental challenge is that AI represents not merely a new tool within existing financial structures but a transformation of those structures themselves. Addressing this challenge requires not only technical expertise in AI but also deep understanding of financial markets, institutional economics, human behavior, and the broader social implications of automated decision-making. It demands unprecedented cooperation between financial institutions, technology companies, regulators, academic researchers, and civil society organizations. Most fundamentally, it requires recognizing that the question is not whether to regulate AI in financial markets but how to do so in ways that maximize benefits while minimizing risks – a challenge that will define financial governance for decades to come.

Minh họa quản lý AI trong thị trường tài chính toàn cầu với mạng lưới kết nối các tổ chức

Questions 27-31

Matching Headings

The passage has nine paragraphs (A-I).

Choose the correct heading for paragraphs B-F from the list of headings below.

List of Headings:
i. The challenge of international regulatory coordination
ii. Traditional definitions of systemic risk
iii. Security vulnerabilities in AI financial systems
iv. The problem of technological homogeneity
v. Embedded regulation as a potential solution
vi. Model risk in the age of AI
vii. The need for adaptive regulatory frameworks
viii. Transparency requirements across jurisdictions
ix. The complexity of neural network validation

27. Paragraph B
28. Paragraph C
29. Paragraph D
30. Paragraph E
31. Paragraph F

Questions 32-36

Complete the sentences below.

Choose NO MORE THAN THREE WORDS from the passage for each answer.

32. After the 2008 crisis, regulators designated certain institutions as __ subject to enhanced oversight.

33. The BIS paper suggests that AI could increase __ during periods of market stress.

34. The passage states that deep neural networks contain parameters whose relationships are __ to human analysis.

35. The European Union’s approach to AI regulation emphasizes __ and protection of individual rights.

36. Some scholars propose using __ as controlled environments for testing AI applications before wider implementation.

Questions 37-40

Do the following statements agree with the claims of the writer in Passage 3?

Write:

• YES if the statement agrees with the claims of the writer
• NO if the statement contradicts the claims of the writer
• NOT GIVEN if it is impossible to say what the writer thinks about this

37. The 2008 financial crisis was primarily caused by AI systems.

38. Similar AI models across multiple institutions could create correlated behavior that amplifies market movements.

39. Embedded regulation is the only effective approach to governing AI in financial markets.

40. Future AI regulation will likely need to evolve more rapidly than traditional regulatory approaches.

---

Answer Keys – Đáp Án

PASSAGE 1: Questions 1-13

1. TRUE
2. FALSE
3. NOT GIVEN
4. NOT GIVEN
5. TRUE
6. NOT GIVEN
7. transaction times
8. arbitrage
9. emotional decision-making
10. kill switches
11. B
12. C
13. B

PASSAGE 2: Questions 14-26

14. C
15. B
16. B
17. C
18. C
19. credit scoring
20. right to explanation
21. algorithmic bias
22. diverse training data
23. algorithmic audits
24. NO
25. YES
26. NO

PASSAGE 3: Questions 27-40

27. ii
28. iv
29. vi
30. ix
31. iii
32. systemically important financial institutions / SIFIs
33. herding behavior
34. mathematically intractable
35. precautionary principles
36. regulatory sandboxes
37. NO
38. YES
39. NO
40. YES

---

Giải Thích Đáp Án Chi Tiết

Passage 1 – Giải Thích

Câu 1: TRUE

• Dạng câu hỏi: True/False/Not Given
• Từ khóa: algorithmic trading, more than half, equity trading, United States
• Vị trí trong bài: Đoạn 2, dòng 1-3
• Giải thích: Bài viết nói rõ “estimates suggest that automated systems are responsible for 60-75% of all equity trading in the United States”. Con số 60-75% rõ ràng lớn hơn 50% (more than half), do đó câu này TRUE. Paraphrase: “more than half” = “60-75%”, “algorithmic trading” = “automated systems”.

Câu 2: FALSE

• Dạng câu hỏi: True/False/Not Given
• Từ khóa: high-frequency trading, hold positions, several hours
• Vị trí trong bài: Đoạn 3, dòng 2-4
• Giải thích: Bài viết khẳng định “holding positions for extremely short periods – sometimes just seconds or milliseconds”. Điều này mâu thuẫn trực tiếp với “several hours” trong câu hỏi. Paraphrase: “several hours” vs “seconds or milliseconds” là sự đối lập rõ ràng.

Câu 3: NOT GIVEN

• Dạng câu hỏi: True/False/Not Given
• Từ khóa: Flash Crash, permanent losses, most investors
• Vị trí trong bài: Đoạn 5
• Giải thích: Bài viết đề cập đến Flash Crash và việc thị trường “recovering most of the loss”, nhưng không đề cập đến việc có hay không có permanent losses cho investors. Thông tin này không có trong bài.

Câu 4: NOT GIVEN

• Dạng câu hỏi: True/False/Not Given
• Từ khóa: Navinder Sarao, convicted, spoofing algorithms
• Vị trí trong bài: Đoạn 6, dòng 3-5
• Giải thích: Bài viết chỉ nói Sarao “was arrested for allegedly using”, không đề cập đến việc ông ta có bị convicted (kết án) hay không. “Arrested” và “allegedly” khác với “convicted”.

Câu 5: TRUE

• Dạng câu hỏi: True/False/Not Given
• Từ khóa: algorithmic trading, reduced, difference between buying and selling prices
• Vị trí trong bài: Đoạn 4, dòng 5-7
• Giải thích: Bài viết nói “The competition among algorithmic traders has narrowed bid-ask spreads (the difference between buying and selling prices)”. Paraphrase: “narrowed” = “reduced”, “bid-ask spreads” = “difference between buying and selling prices”.

Câu 7: transaction times

• Dạng câu hỏi: Sentence Completion
• Từ khóa: trading firms, spend millions, reduce, microseconds
• Vị trí trong bài: Đoạn 2, dòng 6-8
• Giải thích: Câu gốc: “trading firms invest millions of dollars in technology to reduce transaction times by mere microseconds”. Đáp án chính xác là “transaction times”.

Câu 8: arbitrage

• Dạng câu hỏi: Sentence Completion
• Từ khóa: profiting, brief price differences, exchanges
• Vị trí trong bài: Đoạn 3, dòng 6-8
• Giải thích: Bài viết giải thích “This practice, known as arbitrage” sau khi mô tả việc mua bán trên các sàn khác nhau để thu lợi từ chênh lệch giá.

Câu 11: B

• Dạng câu hỏi: Multiple Choice
• Từ khóa: main advantage, market liquidity
• Vị trí trong bài: Đoạn 4, dòng 2-5
• Giải thích: Bài viết định nghĩa market liquidity là “the ease with which assets can be bought and sold without causing major price changes” và giải thích lợi ích là “making it easier for investors to execute trades quickly”. Đáp án B paraphrase ý này chính xác nhất.

Câu 13: B

• Dạng câu hỏi: Multiple Choice
• Từ khóa: regulating algorithmic trading
• Vị trí trong bài: Đoạn cuối, câu cuối
• Giải thích: Bài viết kết luận “regulating this rapidly evolving technology remains an ongoing challenge, requiring continuous adaptation”. Đáp án B (“continuously evolving challenge”) paraphrase chính xác ý này.

Chiến lược làm bài IELTS Reading hiệu quả với kỹ thuật xác định từ khóa và scanning

Passage 2 – Giải Thích

Câu 14: C

• Dạng câu hỏi: Multiple Choice
• Từ khóa: machine learning differs, traditional statistical models
• Vị trí trong bài: Đoạn 2, dòng 1-4
• Giải thích: Bài viết nói rõ machine learning “enables computers to learn from data without being explicitly programmed with rules” và “Instead of following predetermined instructions, these systems identify patterns”. Đáp án C (“identifies patterns without predetermined rules”) tóm tắt chính xác sự khác biệt này.

Câu 15: B

• Dạng câu hỏi: Multiple Choice
• Từ khóa: thin files
• Vị trí trong bài: Đoạn 3, dòng 4-6
• Giải thích: Bài viết định nghĩa “thin files” là “those with limited traditional credit data, including many young people and recent immigrants”. Đáp án B paraphrase chính xác.

Câu 16: B

• Dạng câu hỏi: Multiple Choice
• Từ khóa: black box problem
• Vị trí trong bài: Đoạn 5, dòng 2-6
• Giải thích: Bài viết giải thích black box là các mô hình mà “Even their creators cannot fully explain how they arrive at specific decisions” và “providing such explanations becomes extremely difficult”. Đáp án B (“difficulty of explaining how algorithms make decisions”) chính xác.

Câu 17: C

• Dạng câu hỏi: Multiple Choice
• Từ khóa: proxy variables
• Vị trí trong bài: Đoạn 7, dòng 3-6
• Giải thích: Bài viết giải thích proxy variables là “other factors that correlate with protected characteristics” và đưa ví dụ zip code có thể “serve as a proxy for race”, dẫn đến discrimination. Đáp án C chính xác.

Câu 19: credit scoring

• Dạng câu hỏi: Summary Completion
• Từ khóa: improved, companies like Upstart
• Vị trí trong bài: Đoạn 3-4
• Giải thích: Đoạn 3 bắt đầu với “Credit scoring provides a compelling example” và đoạn 4 nói về Upstart và các cải thiện. Context phù hợp với “improved credit scoring”.

Câu 20: right to explanation

• Dạng câu hỏi: Summary Completion
• Từ khóa: GDPR, Europe, explain automated decisions
• Vị trí trong bài: Đoạn 6, dòng 1-2
• Giải thích: Câu gốc: “The European Union’s General Data Protection Regulation (GDPR)… includes a ‘right to explanation’ for automated decisions”. Đáp án chính xác.

Câu 24: NO

• Dạng câu hỏi: Yes/No/Not Given
• Từ khóa: machine learning models, always more accurate
• Vị trí trong bài: Đoạn 2-3
• Giải thích: Bài viết nói machine learning “often” có superior accuracy và đưa ví dụ cải thiện, nhưng không bao giờ nói “always”. Câu hỏi sử dụng từ tuyệt đối “always” không đúng với quan điểm tác giả. Đáp án là NO.

Câu 25: YES

• Dạng câu hỏi: Yes/No/Not Given
• Từ khóa: Apple Card investigation, real concerns, gender bias
• Vị trí trong bài: Đoạn 8, dòng 1-4
• Giải thích: Bài viết đề cập Apple Card “faced allegations of gender bias” và “Regulators investigated these claims, highlighting the challenges”. Việc regulators điều tra và highlighting challenges cho thấy đây là “real concerns”. Đáp án YES.

Câu 26: NO

• Dạng câu hỏi: Yes/No/Not Given
• Từ khóa: most financial institutions, successfully resolved, transparency issues
• Vị trí trong bài: Đoạn 6, 9-11
• Giải thích: Bài viết nói việc reconciling legal requirements với machine learning complexity “remains an ongoing struggle” và các regulators “are developing responses”. Điều này cho thấy vấn đề chưa được resolved successfully. Đáp án NO.

Passage 3 – Giải Thích

Câu 27: ii (Traditional definitions of systemic risk)

• Dạng câu hỏi: Matching Headings
• Vị trí trong bài: Paragraph B (đoạn 2)
• Giải thích: Đoạn này định nghĩa “systemic risk”, đưa ví dụ 2008 crisis và giải thích cách regulators phản ứng với traditional understanding về systemic risk thông qua việc chỉ định SIFIs. Heading ii phù hợp nhất.

Câu 28: iv (The problem of technological homogeneity)

• Dạng câu hỏi: Matching Headings
• Vị trí trong bài: Paragraph C (đoạn 3)
• Giải thích: Đoạn này tập trung vào “technological homogeneity” và “algorithmic monoculture” – việc nhiều tổ chức sử dụng AI models tương tự tạo ra correlated behavior. Heading iv chính xác.

Câu 29: vi (Model risk in the age of AI)

• Dạng câu hỏi: Matching Headings
• Vị trí trong bài: Paragraph D (đoạn 4-5)
• Giải thích: Đoạn này thảo luận “model risk” và đưa ví dụ về các model failures trong quá khứ, sau đó giải thích tại sao model risk trở nên nghiêm trọng hơn với AI. Heading vi phù hợp.

Câu 32: systemically important financial institutions / SIFIs

• Dạng câu hỏi: Sentence Completion
• Từ khóa: 2008 crisis, regulators designated, enhanced oversight
• Vị trí trong bài: Đoạn 2, dòng 4-5
• Giải thích: Câu gốc: “Regulators responded by designating certain institutions as ‘systemically important financial institutions (SIFIs)’ subject to enhanced oversight”. Đáp án có thể là cụm đầy đủ hoặc viết tắt.

Câu 33: herding behavior

• Dạng câu hỏi: Sentence Completion
• Từ khóa: BIS paper, AI could increase, market stress
• Vị trí trong bài: Đoạn 4, dòng 2-4
• Giải thích: Trích dẫn từ BIS paper: “there is a risk that herding behavior could increase, particularly during periods of stress”. Đáp án chính xác là “herding behavior”.

Câu 37: NO

• Dạng câu hỏi: Yes/No/Not Given
• Từ khóa: 2008 financial crisis, primarily caused, AI systems
• Vị trí trong bài: Đoạn 2
• Giải thích: Bài viết đề cập 2008 crisis liên quan đến “collapse of Lehman Brothers” và institutional problems, không hề đề cập AI là nguyên nhân. Thực tế, bài viết sử dụng 2008 như ví dụ về traditional systemic risk, trước khi AI trở nên phổ biến. Đáp án NO.

Câu 38: YES

• Dạng câu hỏi: Yes/No/Not Given
• Từ khóa: similar AI models, correlated behavior, amplifies market movements
• Vị trí trong bài: Đoạn 3, dòng 3-7
• Giải thích: Bài viết nói rõ similar AI models “may inadvertently create correlated behavior” và “simultaneously recommend selling similar assets during market stress, accelerating price declines”. Đây chính xác là ý “amplify market movements”. Đáp án YES.

Câu 40: YES

• Dạng câu hỏi: Yes/No/Not Given
• Từ khóa: future AI regulation, evolve more rapidly, traditional regulatory approaches
• Vị trí trong bài: Đoạn 12, dòng 2-6
• Giải thích: Bài viết nói “regulatory frameworks for AI in financial markets will likely need to become more adaptive” và giải thích “temporal mismatch between regulatory and technological change” yêu cầu regulation phải “evolve alongside the technology”. Đáp án YES.

---

Từ Vựng Quan Trọng Theo Passage

Passage 1 – Essential Vocabulary

Từ vựng	Loại từ	Phiên âm	Nghĩa tiếng Việt	Ví dụ từ bài	Collocation
algorithmic trading	n	/ˌælɡəˈrɪðmɪk ˈtreɪdɪŋ/	giao dịch thuật toán	Algorithmic trading accounted for less than 10%	automated/high-frequency algorithmic trading
predefined rules	n	/ˌpriːdɪˈfaɪnd ruːlz/	quy tắc được định sẵn	execute trades based on predefined rules	follow/implement predefined rules
high-frequency trading	n	/haɪ ˈfriːkwənsi ˈtreɪdɪŋ/	giao dịch tần suất cao	HFT represents the most sophisticated form	engage in/conduct high-frequency trading
market liquidity	n	/ˈmɑːrkɪt lɪˈkwɪdəti/	tính thanh khoản thị trường	significantly improved market liquidity	enhance/provide/maintain market liquidity
bid-ask spread	n	/bɪd æsk spred/	chênh lệch giá mua-bán	narrowed bid-ask spreads	wide/narrow/tight bid-ask spread
arbitrage	n	/ˈɑːrbɪtrɑːʒ/	kinh doanh chênh lệch giá	This practice, known as arbitrage	exploit/engage in arbitrage
market volatility	n	/ˈmɑːrkɪt ˌvɒləˈtɪləti/	biến động thị trường	introduced new risks including market volatility	increased/high/extreme market volatility
Flash Crash	n	/flæʃ kræʃ/	sụp đổ chớp nhoáng	the Flash Crash of May 6, 2010	experience/trigger a flash crash
spoofing	n	/ˈspuːfɪŋ/	giả mạo lệnh giao dịch	illegal practices such as spoofing	engage in/detect spoofing
risk controls	n	/rɪsk kənˈtrəʊlz/	kiểm soát rủi ro	require firms to implement risk controls	establish/maintain/implement risk controls
kill switch	n	/kɪl swɪtʃ/	công tắc khẩn cấp	automatic kill switches that can halt trading	activate/implement a kill switch
transaction costs	n	/trænˈzækʃən kɒsts/	chi phí giao dịch	reduced transaction costs for many participants	lower/reduce/minimize transaction costs

Passage 2 – Essential Vocabulary

Từ vựng	Loại từ	Phiên âm	Nghĩa tiếng Việt	Ví dụ từ bài	Collocation
risk assessment	n	/rɪsk əˈsesmənt/	đánh giá rủi ro	AI is transforming risk assessment	conduct/perform/carry out risk assessment
machine learning	n	/məˈʃiːn ˈlɜːrnɪŋ/	học máy	machine learning algorithms are handling these tasks	apply/deploy/use machine learning
creditworthiness	n	/ˈkredɪtwɜːrðinəs/	khả năng tín dụng	assessing the creditworthiness of loan applicants	evaluate/assess/determine creditworthiness
training data	n	/ˈtreɪnɪŋ ˈdeɪtə/	dữ liệu huấn luyện	identify patterns in training data	collect/use/process training data
alternative data	n	/ɔːlˈtɜːrnətɪv ˈdeɪtə/	dữ liệu thay thế	alternative data sources such as social media	leverage/incorporate alternative data
credit scoring	n	/ˈkredɪt ˈskɔːrɪŋ/	chấm điểm tín dụng	Credit scoring provides a compelling example	improve/enhance credit scoring
algorithmic opacity	n	/ˌælɡəˈrɪðmɪk əʊˈpæsəti/	sự mờ đục của thuật toán	The first concern involves algorithmic opacity	address/reduce algorithmic opacity
black box	n	/blæk bɒks/	hộp đen (không giải thích được)	function as black boxes	operate as/function as a black box
neural network	n	/ˈnjʊərəl ˈnetwɜːrk/	mạng nơ-ron	using deep learning techniques with neural networks	train/deploy/implement neural networks
algorithmic bias	n	/ˌælɡəˈrɪðmɪk ˈbaɪəs/	thiên vị thuật toán	A second major concern is algorithmic bias	address/mitigate/reduce algorithmic bias
proxy variable	n	/ˈprɒksi ˈveəriəbl/	biến đại diện	discriminate through proxy variables	use/identify/eliminate proxy variables
fairness constraint	n	/ˈfeənəs kənˈstreɪnt/	ràng buộc công bằng	advocate for fairness constraints	implement/apply/enforce fairness constraints
algorithmic audit	n	/ˌælɡəˈrɪðmɪk ˈɔːdɪt/	kiểm toán thuật toán	Regular algorithmic audits can catch problems	conduct/perform/require algorithmic audits
right to explanation	n	/raɪt tuː ˌekspləˈneɪʃən/	quyền được giải thích	GDPR includes a right to explanation	guarantee/provide/ensure right to explanation
democratize access	v	/dɪˈmɒkrətaɪz ˈækses/	dân chủ hóa việc tiếp cận	could democratize access to credit	help/enable/work to democratize access

Passage 3 – Essential Vocabulary

Từ vựng	Loại từ	Phiên âm	Nghĩa tiếng Việt	Ví dụ từ bài	Collocation
systemic risk	n	/sɪˈstemɪk rɪsk/	rủi ro hệ thống	understanding and mitigating systemic risk	pose/create/mitigate systemic risk
systemic architecture	n	/sɪˈstemɪk ˈɑːrkɪtektʃər/	kiến trúc hệ thống	alters the systemic architecture of global finance	transform/change systemic architecture
regulatory paradigm	n	/ˈreɡjələtəri ˈpærədaɪm/	mô hình quản lý	transcend traditional regulatory paradigms	shift/change/establish regulatory paradigms
contagion effect	n	/kənˈteɪdʒən ɪˈfekt/	hiệu ứng lây lan	created a contagion effect that threatened	trigger/cause/create a contagion effect
technological homogeneity	n	/ˌteknəˈlɒdʒɪkəl ˌhɒmədʒəˈniːəti/	sự đồng nhất công nghệ	AI introduces technological homogeneity	lead to/create/increase technological homogeneity
algorithmic monoculture	n	/ˌælɡəˈrɪðmɪk ˈmɒnəkʌltʃər/	đơn canh thuật toán	This algorithmic monoculture could cause	create/result in algorithmic monoculture
herding behavior	n	/ˈhɜːrdɪŋ bɪˈheɪvjər/	hành vi bầy đàn	risk that herding behavior could increase	exhibit/demonstrate/increase herding behavior
model risk	n	/ˈmɒdl rɪsk/	rủi ro mô hình	Model risk takes on systemic dimensions	assess/manage/mitigate model risk
deep neural network	n	/diːp ˈnjʊərəl ˈnetwɜːrk/	mạng nơ-ron sâu	Deep neural networks contain millions of parameters	train/deploy/use deep neural networks
mathematically intractable	adj	/ˌmæθəˈmætɪkli ɪnˈtræktəbl/	không thể giải quyết bằng toán học	relationships are mathematically intractable	remain/prove mathematically intractable
adversarial machine learning	n	/ˌædvəˈseəriəl məˈʃiːn ˈlɜːrnɪŋ/	học máy đối kháng	Adversarial machine learning can deceive AI	use/apply adversarial machine learning
regulatory arbitrage	n	/ˈreɡjələtəri ˈɑːrbɪtrɑːʒ/	kinh doanh chênh lệch quy định	creating regulatory arbitrage opportunities	exploit/engage in regulatory arbitrage
precautionary principle	n	/prɪˈkɔːʃənəri ˈprɪnsəpl/	nguyên tắc phòng ngừa	European approaches emphasize precautionary principles	apply/adopt/follow precautionary principles
regulatory fragmentation	n	/ˈreɡjələtəri ˌfræɡmenˈteɪʃən/	sự phân mảnh quy định	regulatory fragmentation creates substantial challenges	lead to/increase regulatory fragmentation
embedded regulation	n	/ɪmˈbedɪd ˌreɡjuˈleɪʃən/	quy định nhúng	advocate for embedded regulation	implement/develop embedded regulation
RegTech	n	/ˈreɡtek/	công nghệ quản lý	This approach, sometimes called RegTech	deploy/utilize/develop RegTech
regulatory sandbox	n	/ˈreɡjələtəri ˈsændbɒks/	hộp cát quản lý	propose regulatory sandboxes for testing	establish/create/participate in regulatory sandboxes
outcome-based standard	n	/ˈaʊtkʌm beɪst ˈstændərd/	tiêu chuẩn dựa trên kết quả	focus on outcome-based standards	adopt/implement outcome-based standards

Bộ từ vựng IELTS Reading chuyên ngành tài chính và công nghệ AI

---

Kết Bài

Chủ đề “Regulating AI in financial markets” không chỉ phản ánh xu hướng công nghệ hiện đại mà còn là một trong những chủ đề phức tạp và đa chiều thường xuất hiện trong IELTS Reading. Qua bộ đề thi mẫu này, bạn đã được trải nghiệm đầy đủ ba cấp độ khó từ cơ bản đến nâng cao, bao quát các dạng câu hỏi quan trọng nhất trong kỳ thi thực tế.

Ba passages đã cung cấp góc nhìn toàn diện về AI trong tài chính: từ algorithmic trading và các rủi ro thị trường cơ bản (Passage 1), đến machine learning trong đánh giá rủi ro với các thách thức về công bằng và minh bạch (Passage 2), cho đến những vấn đề phức tạp về rủi ro hệ thống và khung pháp lý quốc tế (Passage 3). Mỗi passage không chỉ giúp bạn rèn luyện kỹ năng đọc hiểu mà còn mở rộng kiến thức về một lĩnh vực quan trọng của xã hội đương đại.

Đáp án chi tiết kèm giải thích đã chỉ ra cách xác định thông tin trong bài, nhận biết paraphrase, và áp dụng các chiến lược làm bài hiệu quả. Bộ từ vựng chuyên ngành với hơn 40 từ quan trọng sẽ là nền tảng vững chắc không chỉ cho phần Reading mà còn cho Writing và Speaking khi bạn gặp các chủ đề liên quan đến technology và finance.

Hãy luyện tập đề thi này nhiều lần, phân tích kỹ các câu trả lời sai để hiểu nguyên nhân, và ghi nhớ từ vựng trong ngữ cảnh thực tế. Với sự kiên trì và phương pháp đúng đắn, bạn hoàn toàn có thể đạt được band điểm Reading mục tiêu. Như các nghiên cứu về psychological effects of social isolation cho thấy, việc duy trì động lực và kỷ luật trong học tập là chìa khóa để thành công. Đồng thời, tương tự như cách how digital payments are transforming global remittances, công nghệ AI đang thay đổi cách thức hoạt động của nhiều lĩnh vực, và việc hiểu rõ những thay đổi này sẽ giúp bạn tự tin hơn với các chủ đề học thuật trong IELTS. Chúc bạn ôn tập hiệu quả và đạt kết quả cao trong kỳ thi sắp tới!