NBER Working Paper No. 35756 · Key Findings

The data behind the MRT property-purchase study

Figures from Piskorski, Seru, Zhang and Zhao (September 2026). This is a working paper; PSD is reviewing the methodology.

1–2 yrs before public announcement peak window for excess purchases
12% avg annualised unlevered return on purchases classified as "informed"
S$147k est. additional annual earnings per informed purchase, vs civil servant avg salary
S$132m relatives' excess purchases gross target-property acquisition value
Civil servant vs control purchase rate near planned MRT stations, by timing window

Indexed to 100 = typical purchase rate. Directional illustration of the paper's findings; exact magnitudes vary by model specification.

Civil servants Control group
Annualised excess return vs market benchmark, by buyer group

Civil servants' "informed" purchases outperform the market benchmark by 2.4 pp (unlevered) and 16.3 pp (levered). Neither real estate agents nor board members show a comparable excess.

Pre-2011

Pattern detectable

Excess pre-announcement purchasing by civil servants and co-registered relatives. Concentrated 1–2 years before station announcement.

2011–12

Enforcement tightened:
SFA Amendment Act & CPIB reforms

Post-2012

Pattern disappears

Excess purchasing no longer detectable. Relatives' purchases also cease. No evidence of substitution between channels.

Source: NBER Working Paper No. 35756 (Piskorski, Seru, Zhang, Zhao, September 2026). Timing-window chart is directional and illustrative of the paper's reported patterns. This paper has not been peer-reviewed and PSD has not accepted its findings.

This article summarises NBER Working Paper No. 35756. This is a working paper — not peer-reviewed — and the NBER Board has not reviewed it as an official NBER publication. PSD has not accepted its findings and is currently reviewing the authors’ data and methodology. No individual civil servant is identified in the paper. The statistical findings do not by themselves establish wrongdoing by any individual. AI was used to assist in summarising the 86-page paper — please read the full paper before drawing conclusions.

A new academic paper has found that Singapore civil servants purchased private residential properties near planned MRT station locations at significantly higher rates than comparable private citizens — and did so in the one-to-two-year window before those station locations were publicly announced. The study, published by the National Bureau of Economic Research (NBER) in September 2026, has prompted Singapore’s Public Service Division (PSD) to announce a review of the paper’s data and methodology.

The findings land against a backdrop of global interest in Singapore’s governance model. For decades, Singapore has been held up as proof that corruption can be beaten. The story is well known: a scrappy post-independence government, rampant graft, Lee Kuan Yew, the Corrupt Practices Investigation Bureau (CPIB), high civil service salaries, and the slow, deliberate construction of a culture of integrity. By the 1990s, Singapore sat at the top of every governance ranking. The question scholars then started asking was: had the country finally broken free? Could the norms sustain themselves without constant enforcement?

Four economists — Tomasz Piskorski of Columbia, Amit Seru and Chun Zhao of Stanford, and Jian Zhang of HKU — decided to answer that question empirically. Their method was elegant: link the universe of private property transactions in Singapore to the identities of civil servants, then look at what happened in the years before new MRT station locations were publicly announced.

The answer they found is uncomfortable.

What did the NBER study find about MRT property purchases?

New MRT stations create real estate value. A property near a planned station that nobody knows about yet is worth considerably more once the announcement is made. Anyone who knows the station location in advance — and buys before the announcement — stands to profit. That information is non-public. If the pattern identified by the researchers did result from officials acting on confidential government information, it would raise serious questions about public-service integrity and the handling of official information. The paper’s statistical analysis, however, does not by itself establish that any individual committed an offence.

The researchers assembled housing transaction data from legal/caveat filings at the Registry of Land Titles covering January 1995 to December 2019, then cross-referenced buyers with civil servant identities collected across six cohorts from the Singapore Government Directory Interactive (SGDI) between 2007 and 2022. They also used GIS tools to identify which properties fell within one kilometre of each planned MRT station.

The pattern is clear: Singapore civil servants purchased properties near planned (but unannounced) MRT stations at significantly higher rates than a matched control group of observationally similar private citizens. The difference is concentrated in the one-to-two-year window before public announcement — precisely when the information was most valuable and before prices reflected the coming station.

To rule out the possibility that civil servants are simply better at processing public information, the authors compare them against real estate agents and corporate board members — two groups with strong incentives and high information-processing sophistication. Neither group shows comparable pre-announcement elevation. The advantage appears to be access to non-public information, not analytical skill.

“Civil servants disproportionately purchase homes near planned — but not yet publicly announced — stations, with the effect concentrated one to two years before public announcements.”

— Piskorski, Seru, Zhang & Zhao, NBER Working Paper No. 35756

Which civil servants and which agencies showed the pattern?

The excess pre-announcement purchasing is not uniform across the civil servants in the dataset. Using LLM-assisted classification of civil service ranks, the authors find the pattern is concentrated among mid-level officials — not senior staff and not junior staff. The authors’ explanation is that mid-level officials may have had sufficient financial resources to act while facing less expected scrutiny and reputational cost than senior officials; junior officials, meanwhile, may have faced greater resource constraints.

Agencies involved in rail-line planning show higher rates of the excess-purchasing pattern than those further from the planning process — which the authors interpret as consistent with greater access to non-public MRT location information.

The researchers also construct a group they describe as “relatives” — individuals co-registered at the same residential address as identified civil servants. These co-registered individuals also exhibited elevated pre-announcement purchasing patterns and earned returns above the market benchmark. The authors estimate approximately SGD 132 million in gross target-property acquisition value associated with this connected group’s excess pre-announcement purchases.

How much did civil servants earn from these MRT-adjacent purchases?

Civil servants do not pay a price premium when buying — sellers don’t know the buyer’s informational advantage. But when they sell, the returns are significant. On an unlevered basis, purchases the researchers classify as informed outperform the market benchmark by an average of 2.4 percentage points annually. On a levered basis — accounting for typical mortgage financing — the gap widens to 16.3 percentage points. Against average civil servant incomes of roughly SGD 136,600 per year, the excess return represents between 21% and 61% of annual salary on a single transaction.

Buyers in the researchers’ identified group also hold properties for shorter periods than comparable control buyers, which the authors interpret as consistent with investment rather than consumption motives.

Did the pattern stop after enforcement tightened?

Yes — and this is the paper’s most consequential finding. In 2011–2012, Singapore strengthened several elements of its enforcement framework. The Securities and Futures (Amendment) Act 2012 expanded insider trading provisions and enhanced investigative powers for the Monetary Authority of Singapore. The same period saw several high-profile corruption scandals at the Central Narcotics Bureau, the Singapore Civil Defence Force, and the CPIB itself.

After this period, the excess pre-announcement MRT property purchasing pattern is no longer detectable in the data. Relatives’ pre-announcement purchases also disappear. The researchers find no evidence of substitution between the two channels — that is, no sign of officials routing trades through family members after direct purchases became riskier. Both patterns simply cease to be detectable.

Survey data from the Asian Barometer and World Values Survey showed perceived corruption rising and pro-integrity norms weakening around the 2011–2012 period. The researchers argue that their transaction-level evidence is consistent with this survey-measured deterioration reflecting real behavioural change — not merely a shift in perceptions or media coverage. Whether the enforcement changes caused the disappearance of the statistical pattern, or both were responses to a common underlying factor, is a question the study cannot definitively resolve.

What is the paper’s broader argument about anti-corruption norms?

The researchers frame this not as a scandal exposé but as a test of a fundamental question in governance: can social norms, once established through decades of strict enforcement, become self-sustaining? Can a government eventually step back from active policing and rely on culture to do the work?

Their answer is no — or at least, not yet, and probably not in any jurisdiction less well-governed than Singapore. Even after decades of stringent anti-corruption enforcement in one of the world’s most admired public services, the norms proved fragile. When the deterrence environment softened, behaviour shifted. When deterrence strengthened again, it stopped.

The authors are careful not to draw conclusions about the scale of corruption in Singapore’s civil service, and the numbers they document — while meaningful — should be read against a public sector of approximately 141,000 officers. What the paper establishes is the existence and character of the phenomenon, not its prevalence.

What has PSD said about the NBER paper?

Singapore’s Public Service Division has acknowledged awareness of the paper. According to a Channel NewsAsia report, PSD said it is reviewing the paper’s data and methodology. PSD added that it will refer the matter to the Corrupt Practices Investigation Bureau (CPIB) if there is a material basis to do so.

What this study does — and does not — establish

The study identifies a statistical pattern in transaction records. It does not identify any individual civil servant. It does not establish that any person committed an offence. Its methodology and data have not yet been independently peer-reviewed, and PSD has not accepted its findings.

What the paper does establish — to the satisfaction of its authors — is that the statistical pattern exists in the data, that it is concentrated among mid-level officials and co-registered relatives, that it is associated with excess returns above the market benchmark, and that it ceased to be detectable following enforcement tightening in 2011–2012.

Whether that pattern reflects deliberate use of non-public information, structural advantages that do not involve any breach of duty, data artefacts, or something else entirely is not resolved by the study. The authors offer an interpretation; they acknowledge it is an inference. PSD’s review, and any subsequent CPIB referral, may produce more definitive answers. Or it may not. Readers are encouraged to read the full paper and form their own view.

Sources: NBER Working Paper No. 35756 · Piskorski, Seru, Zhang, Zhao · September 2026 | Channel NewsAsia · Singapore · 2026

Further updates by CNA: NUS academics distance themselves from study on civil servants’ home purchases near unannounced MRT stations · 15 Sep 2026