EC484 Current-Scope Exam Questions Summary

做完之后第一时间去 EC484 Final Review WT 白皮书 总结考点。

EC484 本学期范围内考试题总结

Scan Summary

• 已排除：AT/ 文件夹下所有内容。
• 当前范围依据：只用 Slides/ 下本学期课件判断是否在范围内。
• 当前课件：Slides/ec484-topic0.pdf、Slides/ec484-paneldata.pdf、Slides/ec484-timeseries.pdf、Slides/Bootstrap.pdf、Slides/Causal.pdf、Slides/Limdep.pdf、Slides/Machine.pdf。
• 考试材料：Past Exams/EC484_2018.pdf 到 Past Exams/EC484_2025.pdf、Past Exams/EC484LT_2018.pdf 到 Past Exams/EC484LT_2024.pdf、Sample-EC484-2026-ST.pdf。
• 未重复计入：Past Exams/PastExam_Combination.pdf 是历年卷合集，内容与单年 PDF 重复。
• 说明：本文只总结考试题型和考点，不解题、不生成完整课件笔记。

Current Scope Basis

Current topic	Lecture locator
Topic 0: regression estimators, assumptions A1-A5, GMM, MLE, asymptotics, IV	ec484-topic0 课件页 3-58
Panel data: FE/RE, dynamic panel, panel LDV/GMM	ec484-paneldata 课件页 1-40
Time series: stationarity, ergodicity, ARMA, robust variance, definitions-only special topics	ec484-timeseries 课件页 1-19
Bootstrap: algorithm, bootstrap variance/CI/test/bias/MSE, OLS/GMM bootstrap	Bootstrap 课件页 3-40
Causal inference: ATE/ATT, propensity score/IPW, matching, IV/LATE, DID, RDD, weak IV	Causal 课件页 3-73
Limited dependent variables: MLE tests, binary/ordered/multinomial choice, Tobit/censored/truncated, sample selection	Limdep 课件页 3-82
Machine learning: AIC/CV, ridge, lasso, Lasso IV, partialling-out Lasso	Machine 课件页 4-40

High-Yield Question Families

1. Limited Dependent Variables and MLE

Exam pattern	Past exam evidence	Current lecture location	What the question usually asks
General MLE definition, computation, asymptotic distribution, and Wald/LR/Score tests	EC484_2018 Q3a PDF页 5; EC484_2024 Q3a PDF页 4; Sample 2026 Q1a(i) PDF页 2	Limdep 课件页 3-22; ec484-topic0 课件页 39-50	Define the likelihood/log-likelihood, state MLE, use score/Hessian/sandwich information, or choose two test procedures.
Binary probit/logit and marginal effects	EC484_2022 Q3a-c PDF页 4; EC484_2023 Q3a PDF页 4; Sample 2026 Q1a PDF页 2	Limdep 课件页 23-38; Limdep 课件页 32; ec484-topic0 课件页 47-50	Define probit/logit MLE, derive asymptotic distribution, use delta method for marginal effects, build asymptotic or bootstrap CI, discuss misspecified probit/logit pseudo-true limits.
Ordered probit / ordered latent-variable models	EC484_2018 Q3c PDF页 5; EC484_2021 Q3a-c PDF页 4	Limdep 课件页 39-45	Translate latent thresholds into category probabilities, write log-likelihood, then state MLE asymptotics or CI for expected outcome.
Endogenous regressor in binary/ordered probit	EC484_2019 Q3 PDF页 4; EC484_2021 Q3d PDF页 4	Limdep 课件页 35-38	Recognize endogeneity through correlated latent errors, use control-function/residual inclusion logic, estimate first stage, then include generated residual in probit/ordered probit.
Censored, truncated, Tobit, censored LAD, top-coding	EC484_2020 Q3 PDF页 5; EC484_2023 Q3b PDF页 4; Sample 2026 Q1b PDF页 2	Limdep 课件页 59-71; Limdep 课件页 61-67; Bootstrap 课件页 29-30	Write likelihood for observed mass/continuous parts, handle heteroskedastic or endogenous variants, contrast MLE/NLS/censored LAD, and implement bootstrap tests.

2. Causal Inference, Treatment Effects, RDD, and Weak IV

Exam pattern	Past exam evidence	Current lecture location	What the question usually asks
Potential outcomes, CI/overlap, IPW, ATT/ATE identification	EC484_2018 Q3b PDF页 5; EC484_2024 Q3b PDF页 4; Sample 2026 Q2a PDF页 3	Causal 课件页 3-18	Prove an IPW or regression-identification formula, then suggest parametric or nonparametric estimation.
Doubly robust / augmented IPW style expression	EC484_2024 Q3b(iv) PDF页 4	Causal 课件页 8-17	The building blocks are current: conditional mean plus propensity score. Treat this as in-scope only if the exam explicitly asks for the displayed double-robust identity.
IV for treatment effects, LATE, and weak-IV asymptotics	EC484_2020 Q1 PDF页 2; EC484_2021 Q1c PDF页 2; EC484_2022 Q1d PDF页 2; Sample 2026 Q2b PDF页 3	Causal 课件页 19-33; Causal 课件页 56-73; ec484-topic0 课件页 51-54	Derive IV distribution, explain local-to-zero first stage, construct weak-IV robust confidence sets, or interpret random limits under weak identification.
Regression discontinuity design	Sample 2026 Q2c PDF页 3	Causal 课件页 48-55	Recognize threshold assignment, identify a causal effect at the cutoff, distinguish sharp vs fuzzy RDD, and describe local estimation.
DID / two-way fixed-effect causal interpretation	No direct current-scope past-exam question found in the scanned papers	Causal 课件页 34-47; ec484-paneldata 课件页 15-23	Still current because it is in the slides, but the scanned exam papers did not show a clean standalone DID question.

3. Bootstrap

Exam pattern	Past exam evidence	Current lecture location	What the question usually asks
Bootstrap bias, variance, MSE, CI, and p-value/test	EC484_2018 Q1b(iii) PDF页 3; EC484_2019 Q2d PDF页 3; EC484_2022 Q3b PDF页 4; EC484_2023 Q1c PDF页 2; EC484_2024 Q1a(iv) PDF页 2; Sample 2026 Q1b(ii) PDF页 2	Bootstrap 课件页 4-8; Bootstrap 课件页 29-33; Bootstrap 课件页 34-40	State the resampling algorithm, define bootstrap statistic, use empirical quantiles or bootstrap p-values, and explain whether recentering is needed.
Bootstrap for GMM / overidentification tests	EC484_2019 Q2d PDF页 3; EC484_2023 Q1c PDF页 2	Bootstrap 课件页 34-40; ec484-topic0 课件页 23-25	Recenter moments if the sample moment is not zero at the estimator; use bootstrap distribution for test statistic or MSE.

4. IV, GMM, Moment Estimation, and Misspecification

Exam pattern	Past exam evidence	Current lecture location	What the question usually asks
IV estimator, asymptotic distribution, variance estimator	EC484_2019 Q1a-d PDF页 2; EC484_2020 Q1a-b,d PDF页 2; EC484_2021 Q1a PDF页 2; EC484_2022 Q1a PDF页 2; Sample 2026 Q2b PDF页 3	ec484-topic0 课件页 51-54; Causal 课件页 56-73	Define IV as moment estimator, state LLN/CLT assumptions, derive asymptotic variance, and estimate it consistently.
Two-step GMM, optimal weight, J-test / overidentification	EC484_2018 Q2d PDF页 4; EC484_2019 Q2a-e PDF页 3; EC484_2021 Q1b PDF页 2; EC484_2022 Q2a-b PDF页 3; EC484_2023 Q2a-e PDF页 3; EC484_2024 Q2a-b PDF页 3	ec484-topic0 课件页 23-25; Bootstrap 课件页 37-40; Causal 课件页 58-71	Set up sample moments, choose weight matrix, derive asymptotic distribution, state optimality, and interpret J-test rejection.
Omitted variables, measurement error, and misspecified moments	EC484_2022 Q1b-c PDF页 2; EC484_2024 Q2d-e PDF页 3; ec484-topic0 课件页 55-58	ec484-topic0 课件页 51-58	Compute probability limits under the true DGP instead of the assumed model; decide whether consistency survives.
Local power / one-sided tests in GMM-type settings	EC484_2022 Q2b PDF页 3	Limdep 课件页 11-18; ec484-topic0 课件页 42-50	State test statistic and local alternative, then characterize rejection probability under local drift.

5. Machine Learning and High-Dimensional Regression

Exam pattern	Past exam evidence	Current lecture location	What the question usually asks
Ridge vs Lasso when $k>n$	Sample 2026 Q2d PDF页 3	Machine 课件页 17-30	Suggest two estimators for high-dimensional regression, compare shrinkage, variable selection, bias-variance trade-off, and tuning by cross-validation.
Lasso IV / partialling-out Lasso	No direct past-exam question found, but current slides cover it	Machine 课件页 31-40	Current but not yet strongly represented in the scanned exam papers. Expect conceptual comparison or algorithmic outline rather than long derivation.
AIC/CV/post-selection inference	No direct past-exam question found	Machine 课件页 4-16	Current but not strongly represented in the scanned exam papers.

6. Vassilis Part: Asymptotics, Time Series, and Specialized Regression

Exam pattern	Past exam evidence	Current lecture location	What the question usually asks
$O_p$, $o_p$, convergence in distribution/probability/mean, sample variance order	EC484LT_2018 Q2 PDF页 3; EC484LT_2022 Q1 PDF页 2; EC484_2025 Q1 PDF页 2	ec484-topic0 课件页 42-50	Show stochastic boundedness, improve order when the limit is zero, or derive stochastic order for sample moments.
Uniform maximum / boundary normalization	EC484LT_2019 Q2b PDF页 2; EC484LT_2022 Q1c PDF页 2; EC484_2025 Q1c PDF页 2	ec484-topic0 课件页 42-50	Find $g(n)$ so a nonstandard estimator such as ${\max }_i{x}_i$ has a nondegenerate limiting distribution.
Linear process averages, long-memory order, AR(1)/MA processes	EC484LT_2021 Q1 PDF页 2; EC484LT_2021 Q2c PDF页 3; EC484LT_2022 Q2b-c PDF页 3; EC484LT_2023 Q2b-c PDF页 3; EC484LT_2024 Q2 PDF页 3; EC484_2025 Q2b-c PDF页 3	ec484-timeseries 课件页 7-15; ec484-timeseries 课件页 14-19	Prove LLN/CLT-type results for dependent data, identify stationarity/ergodicity conditions, handle autocovariance decay, or show convergence of quadratic averages.
Trending regressions, heteroskedastic constants, efficient weighted estimators	EC484LT_2018 Q3 PDF页 4; EC484LT_2020 Q3 PDF页 3; EC484LT_2021 Q2-Q3 PDF页 3-4; EC484LT_2024 Q3 PDF页 4; EC484_2025 Q2a PDF页 3	ec484-topic0 课件页 29-38; ec484-topic0 课件页 42-50; ec484-timeseries 课件页 1-6	Work out rate/normalization of LS or weighted LS when regressors/error variances trend with $i$ or $t$.
Constrained LS at boundary	EC484LT_2022 Q3 PDF页 5; EC484LT_2023 Q3 PDF页 5; EC484_2025 Q3 PDF页 5	ec484-topic0 课件页 42-50	Obtain nonstandard limiting distribution when the true parameter is on the boundary of $[0,1]$.
Panel data FE/RE/dynamic panel	No clean standalone past-exam question found in the scanned papers	ec484-paneldata 课件页 15-30	Current because it is in the slides, but not strongly represented in these exam PDFs.

Year-by-Year Current-Scope Index

Source	Current-scope questions to care about	Lower-priority / filtered notes
Sample-EC484-2026-ST.pdf	Q1a probit MLE, marginal effect CI, NLS, misspecified logit; Q1b censored wage MLE and bootstrap test; Q2a ATT identification; Q2b weak IV; Q2c RDD; Q2d ridge/lasso. Sample 2026 PDF页 2-3; Limdep 课件页 23-71; Bootstrap 课件页 29-40; Causal 课件页 3-73; Machine 课件页 17-30	Q2e and Q3 are placeholders for Vassilis part, no actual question text.
EC484_2018.pdf	Q1b(iii) bootstrap bias for LAD; Q2d optimal IV-GMM for probit-style conditional CDF; Q3a MLE/tests; Q3b IPW identity; Q3c ordered probit. EC484_2018 PDF页 2-5; Bootstrap 课件页 31-40; Limdep 课件页 3-45; Causal 课件页 8-18; ec484-topic0 课件页 23-25	Q1a and parts of Q1b on projection/LAD asymptotics are less directly represented in the current slides; keep only if instructor emphasized LAD/linear projection.
EC484_2019.pdf	Q1 IV asymptotics and variance; Q2 one-step vs two-step GMM and bootstrap MSE; Q3 endogenous probit/control function. EC484_2019 PDF页 2-4; ec484-topic0 课件页 23-25; ec484-topic0 课件页 51-54; Limdep 课件页 35-38; Bootstrap 课件页 33-40	Q1e semiparametric partialling-out is adjacent to current ML/semiparametric ideas but not a central current slide topic.
EC484_2020.pdf	Q1 endogenous binary regressor with IV, efficient estimator, generated-regressor/probit first stage; Q3 censored/top-coded model, heteroskedastic censored MLE, censored LAD, endogenous censored model. EC484_2020 PDF页 2,5; ec484-topic0 课件页 51-54; Causal 课件页 56-73; Limdep 课件页 59-71; Limdep 课件页 35-38	Q2 kernel density/CDF/bandwidth is not a standalone current lecture topic; ignore unless nonparametrics is explicitly restored.
EC484_2021.pdf	Q1 trimmed IV/GMM and weak-IV local asymptotics; Q3 ordered probit, CI for conditional mean, semiparametric ordered model, endogenous ordered probit. EC484_2021 PDF页 2,4; ec484-topic0 课件页 23-25; Causal 课件页 56-73; Limdep 课件页 39-45; Limdep 课件页 35-38	Q2 joint kernel density/conditional density is lower priority under current slides.
EC484_2022.pdf	Q1 IV with omitted variable, IV measurement error, weak-IV robust confidence set; Q2 GMM variance/local power; Q3 binary probit/logit misspecification, bootstrap CI, semiparametric binary model. EC484_2022 PDF页 2-4; ec484-topic0 课件页 23-25; ec484-topic0 课件页 51-58; Causal 课件页 56-73; Limdep 课件页 23-38; Bootstrap 课件页 26-40	Q2d nonparametric conditional variance is lower priority unless generic kernel methods are restored.
EC484_2023.pdf	Q1c bootstrap test for linear restrictions; Q2 GMM/MoM/J-test/MLE using uniform distribution; Q3 binary marginal effect and censored/truncated likelihood. EC484_2023 PDF页 2-4; Bootstrap 课件页 29-40; ec484-topic0 课件页 23-25; Limdep 课件页 23-71	Q1a IV quantile regression and Q1b generic kernel regression are not central current-slide topics.
EC484_2024.pdf	Q1a(iv) bootstrap test; Q1c one-step Newton/extremum consistency; Q2 GMM, J-test, misspecification, measurement error; Q3a Pareto MLE consistency; Q3b treatment-effect IPW/augmented IPW identity. EC484_2024 PDF页 2-4; Bootstrap 课件页 29-40; ec484-topic0 课件页 23-25; ec484-topic0 课件页 42-58; Limdep 课件页 3-18; Causal 课件页 8-18	Q1b generic nonparametric derivative is only indirectly related to current RDD/local estimation.
EC484_2025.pdf	Q1 stochastic order/convergence and uniform maximum normalization; Q2 trend regression variance estimator plus linear-process averages; Q3 constrained LS at boundary. EC484_2025 PDF页 2-5; ec484-topic0 课件页 42-50; ec484-timeseries 课件页 7-15	Treat as Vassilis/asymptotic-current, but some details may require lecture notes beyond slide text.
EC484LT_2018.pdf	Q2 $O_p/o_p$ algebra and sample variance consistency/order; Q3 trending heteroskedastic regression and efficient estimator. EC484LT_2018 PDF页 3-4; ec484-topic0 课件页 29-38; ec484-topic0 课件页 42-50	Q1 exponential tails and special convergence results are lower priority unless covered in class.
EC484LT_2019.pdf	Q2 uniform maximum normalization is relevant to the repeated nonstandard-asymptotics pattern. EC484LT_2019 PDF页 2; ec484-topic0 课件页 42-50	Q1 uniform integrability and Q3 spectral-density/mixing CLT are not explicit in current slides; ignore unless instructor says otherwise.
EC484LT_2020.pdf	Q3 trending regression with $\gamma$, convergence in second mean, complete convergence, and asymptotic distribution. EC484LT_2020 PDF页 3; ec484-topic0 课件页 42-50; ec484-timeseries 课件页 1-6	Q1 uniform integrability and Q2 extremes are lower priority under current slide scope.
EC484LT_2021.pdf	Q1 regression with MA/AR errors and GLS; Q2 trend regression variance estimator and AR(1) quadratic average; Q3 heteroskedastic constants and exponential regressor normalization. EC484LT_2021 PDF页 2-4; ec484-topic0 课件页 29-50; ec484-timeseries 课件页 7-15	No major filtered part beyond detail level.
EC484LT_2022.pdf	Q1 stochastic order and uniform maximum normalization; Q2 trend regression variance estimator plus linear-process average; Q3 constrained LS at boundary. EC484LT_2022 PDF页 2-5; ec484-topic0 课件页 42-50; ec484-timeseries 课件页 7-15	No major filtered part beyond detail level.
EC484LT_2023.pdf	Q2 trend regression variance estimator plus linear-process average; Q3 constrained LS at boundary. EC484LT_2023 PDF页 3-5; ec484-topic0 课件页 42-50; ec484-timeseries 课件页 7-15	Q1 uniform integrability is not explicit in current slides; ignore unless class notes add it.
EC484LT_2024.pdf	Q2 long-memory LS order with/without intercept; Q3 heteroskedastic constants and exponential regression normalization. EC484LT_2024 PDF页 3-4; ec484-topic0 课件页 42-50; ec484-timeseries 课件页 7-15	Q1 uniform integrability/WLLN/linear-process sufficient condition is not explicit in current slides; likely lower priority. This PDF was image-based on pages 2-4, so these question summaries were checked from rendered pages.

Things To Ignore Unless Confirmed By Instructor

• Standalone kernel density / kernel CDF / bandwidth derivations from old ST papers: EC484_2020 Q2 PDF页 3-4; EC484_2021 Q2 PDF页 3; EC484_2023 Q1b PDF页 2; EC484_2024 Q1b PDF页 2. Current slides include RDD/local estimation and some semiparametric references, but not a full standalone kernel-density topic. Causal 课件页 48-55; Limdep 课件页 33-34
• Uniform integrability proof questions: EC484LT_2019 Q1 PDF页 2; EC484LT_2020 Q1 PDF页 2; EC484LT_2023 Q1 PDF页 2; EC484LT_2024 Q1 PDF页 2. Current slides discuss LLN/ergodicity/moment caveats, but the exact uniform-integrability theorem is not explicit. ec484-topic0 课件页 26-28; ec484-timeseries 课件页 7-13
• Spectral-density and mixing CLT derivations from older LT papers: EC484LT_2019 Q3 PDF页 3. Current time-series slides cover stationarity/ARMA concepts, but not that full derivation. ec484-timeseries 课件页 7-15
• IV quantile regression and standalone quantile/LAD asymptotics: EC484_2023 Q1a PDF页 2; parts of EC484_2018 Q1 PDF页 2-3. Current slides mention LAD and IV/GMM, but do not make IV quantile regression a central topic. ec484-topic0 课件页 21-25; ec484-topic0 课件页 51-54

Quick Priority List

1. Highest priority: Sample 2026 Q1-Q2 because it directly matches current Otsu slides. Sample 2026 PDF页 2-3; Limdep 课件页 23-71; Causal 课件页 3-73; Machine 课件页 17-30
2. Very high priority: LDV/probit/censored/ordered/endogenous binary questions across EC484_2018, EC484_2019, EC484_2020, EC484_2021, EC484_2022, EC484_2023. Limdep 课件页 23-82
3. Very high priority: IV/GMM/weak-IV questions across EC484_2019 to EC484_2024 and Sample 2026 Q2b. ec484-topic0 课件页 23-25; ec484-topic0 课件页 51-54; Causal 课件页 56-73
4. High priority: bootstrap tests, CI, bias/MSE, especially when attached to LDV/GMM. Bootstrap 课件页 29-40
5. High priority for Vassilis: stochastic order, trend regressions, linear processes, constrained boundary estimators. ec484-topic0 课件页 42-50; ec484-timeseries 课件页 7-15
6. Medium priority: panel-data slides because they are current, but the scanned past exams contain few clean panel-only questions. ec484-paneldata 课件页 1-40