config.yml and set future: false.]]>The foundation of any quantitative strategy is the identification of mispriced assets. In sports, this involves building a model to predict game outcomes ($P_{model}$) and comparing those results to the implied probabilities of market odds ($P_{market}$).When a significant deviation occurs—where your model suggests a higher probability of an outcome than the bookmakers’ odds imply—you have identified “Value.” Much like a quant trader buying an undervalued stock, a sports bettor “buys” the outcome where the market has underestimated the true probability.
The most critical insight in this approach is moving away from the “gambler’s fallacy” of betting on a single game. Instead, we treat sports betting as a high-frequency portfolio strategy.
Independence of Events: Unlike stocks, which often move in correlation due to macroeconomic factors, the result of a Saturday afternoon football match in London is almost entirely independent of one in Madrid.
Volatility Reduction: By placing smaller, calculated bets across hundreds or thousands of independent games, the variance (risk) of the total portfolio decreases.
The Law of Large Numbers: While any single game is a toss-up, a model with a slight “edge” (e.g., 53% accuracy) will almost certainly yield a profit over a large enough sample size.
To turn this brainstorm into a functional strategy, one must apply rigorous Position Sizing. In finance, we might use the Kelly Criterion to determine the optimal amount of capital to risk on each “trade” based on the perceived edge and the probability of success. K = \frac{p(b+1) - 1}{b}
Where:$p$ is the probability of winning. $b$ is the decimal odds minus 1.
Sports betting, when stripped of its emotional and “gambling” context, is simply a data science problem. By utilizing econometrics to find market inefficiencies and applying portfolio diversification to manage volatility, we can transform a high-risk hobby into a disciplined quantitative strategy.
]]>Imagine an application that captures every spark of inspiration, every cultural observation, and every fleeting thought, processing them through an AI that understands you. This is log(life).
Instead of a static guidebook, log(life) creates a dynamic, narrated map of your journey. When we travel, we are often at our most observant. We notice a “new social logic” in a Tokyo subway or a “new economic phenomenon” in a Lagos market. Usually, these insights vanish the moment we board the plane home.
log(life) turns your conversations into a living, breathing version of The Lonely Planet. By logging your spoken reactions to new foods, cultures, and norms, the AI constructs a narrative map of your journey. It’s not just a list of places you went; it’s a record of how those places changed your mind.
Writing a diary requires a discipline most of us eventually lose to the friction of daily life. But while we may stop writing, we never stop speaking.log(life) transforms the burden of journaling into a seamless, ambient experience. By identifying “Activity Clusters”—like turning a complex afternoon debate on labor economics into a milestone of intellectual growth—the app ensures you no longer have to stop to record your history. You simply have to live it. The AI even moves beyond text, evolving past manual habit trackers to infer your well-being from the nuances of your voice and the cadence of your speech.
The most profound value of log(life) emerges at the sunset of a journey. As we age, memory often becomes a sieve—names fade, and details blur. This app serves as a “Wisdom Archive,” a digital vessel that allows us to revisit not just the words, but the specific, vibrant energy of our younger selves. Decades from now, a grandchild won’t just read a dry transcript of your history; they can ask, “What did my grandfather feel on his first day of PhD research?” and hear the actual tremor of excitement in your voice. It offers the next generation the “correct” version of your story, preserved in the high fidelity of truth.
The name log(life) carries a profound dual meaning that bridges two languages. In English, it is a literal command: to “log” your life is to document it. But in Chinese, the word for logarithm is 对数 (duìshù), which translates literally to “The Correct Number.”
There is a beautiful mathematical truth here. If life is a chaotic series of variables, log(life) is the function that solves for the “correct form” of your existence. It distills the exponential noise of daily life into a stable, enduring narrative—one that remains true long after the moment has passed.
]]>What is the objective/measure for the goal of this non-profit?
]]>The first-year PhD courses gathered all these pieces into the same room—and suddenly, they started talking to each other. Slowly and interestingly, I realized that economics isn’t a collection of separate languages. It’s one big conversation. And learning to follow that conversation—across theory, data, policy, and back—is probably the real value of the first-year sequence.
Looking back, I think one thing that could make this journey easier for new students is more intentional interdisciplinary connections in the classroom. Not in the “here’s a random paper we won’t have time to discuss” way, but in the “here’s how this model actually lives in the real world” way.
Below are a few examples of what I wish instructors emphasized more.
Take the classic topics: adverse selection and moral hazard. The first-year micro sequence often teaches them as elegant models—beautiful, clean, and completely detached from messy reality.
But it would be far more illuminating to walk students through the full circle:
For example:
Start with the basic moral hazard model.
Then show a Health IO paper trying to identify moral hazard using real insurance data.
Connect this to the econometrics: what makes moral hazard empirically identifiable? What assumptions do we lean on?
Move to policy: how do governments try to solve this? What failures or tradeoffs do they face?
And finally: how does the original model guide welfare analysis and policy design?
The same can be done with auction theory—connecting abstract bidding strategies to how economists estimate auction models in IO or how governments design real-world spectrum auctions using the exact theory taught in class.
Once you see the loop, you can never unsee it. And that’s a good thing.
Econometrics is arguably the easiest place to connect with empirical papers, but we often limit ourselves to old classics. It’s more exciting—and more realistic—to show students how frontier empirical papers push the boundaries of identification.
The causal inference and counterfactual frameworks from the first-year sequence show up everywhere:
labor economics
development
IO
public finance
health
political economy
What would really help first-years is more time spent on assumptions. Not just “state the assumptions and prove the theorem,” but questions like:
Does this empirical paper actually satisfy the assumptions from lecture?
What happens if the assumptions fail?
How would you fix the problem?
Is the estimator still meaningful, or does everything fall apart?
Empirical econometrics becomes a lot more engaging—and honest—when students see that identification is rarely clean and always debatable.
The DSGE portion of macro is famously heavy: Bellman equations, Euler equations, Lagrangians—the usual suspects. We spend weeks solving models by hand, which is important, but ultimately incomplete.
What’s often missing is a clearer sense of what can and cannot be identified from macro data.
For example:
Which parameters in a DSGE model are actually pinned down by data?
Which ones will always be “calibrated”?
How much information do aggregate time series actually contain?
What are the trade-offs between Bayesian estimation, GMM, and indirect inference?
Why is identification in macro so challenging in the first place?
This perspective makes macro feel less like a black box and more like a field with real data limitations and modeling constraints.
The first-year PhD sequence is often described as a “rite of passage,” but maybe it’s better understood as an exercise in synthesis. We enter the program seeing economics as separate domains; we leave the first year realizing that the walls are thin and the doors between fields are wide open.
If instructors can help students see these connections earlier—through interdisciplinary examples, real research, and honest discussions of data and identification—the first-year experience could feel less like survival training and more like intellectual exploration.
And maybe, just maybe, the dots would connect a little sooner.
]]>This isn’t a guide on how to deceive people. Instead, think of it as a defense manual. 🛡️ In our daily lives, we can’t run perfect scientific experiments. If you send your child to music class and their math scores drop, you have to decide whether to continue based on a hunch about the causal link. This uncertainty creates an opportunity for others—advertisers, politicians, or anyone with an agenda—to intentionally create a false sense of causality to achieve their goals.
Here are two of the most common tricks they use.
This trick involves strategically assigning the most promising candidates to the most promising situations, then taking all the credit when things go well. It’s not about creating success; it’s about choosing a winner in advance.
A classic example comes from political appointments. Imagine there are two open mayoral positions and two candidates. In many systems, a mayor’s career depends on the city’s economic growth. But is a mayor truly responsible for a city’s GDP growth? It could just be luck or a shift in the global economy. To game this system, a well-connected or favored candidate might be deliberately assigned to a city that’s already primed for success—perhaps it has healthy public finances or a new industrial park is about to open. A few years later, when the city inevitably thrives, the candidate is hailed as a visionary and promoted. The success was baked in from the start.
We also see this in pop culture. In the Korean drama Sky Castle, a legendary tutor has a perfect record of getting her students into top universities. Desperate parents pay her astronomical fees for this “guaranteed” success. Her secret? She only accepts one or two of the absolute best students each year—students who were likely to succeed with any tutor. Her perfect record is a self-fulfilling prophecy: as her fame grows, she gets to pick from an even more elite pool of applicants, making her “causal” effect look even stronger.
This strategy involves manipulating what information you see. By either hiding the failures or hiding the real cause, a manipulator can craft a very convincing—but completely false—narrative.
Hiding the Failures
This is the art of only showing the winners. By making the losers invisible, a mediocre or even random outcome can be made to look like a stroke of genius. This is known as survivorship bias.
The best illustration is a classic stock market scam. A scammer gets 1,000 phone numbers. He tells 500 that the market will rise tomorrow and tells the other 500 it will fall. The next day, the market rises. He discards the 500 who got the wrong prediction and repeats the process with the remaining group. After a week, a handful of people will have received six or seven “perfect” predictions in a row. To them, the scammer looks like an oracle with insider information. Now that a false causal link (“he knows the future”) is established, the scammer can start charging for his “advice.”
You also hear this logic when people defend unhealthy habits. The argument, “My grandfather smoked his whole life and lived to be 90,” presents a single, lucky survivor as evidence. It conveniently ignores the thousands of others who weren’t so lucky and can’t speak up because they are no longer with us.
Hiding the Real Cause
Sometimes, the outcome is real, but its cause is misattributed. This is done by deliberately hiding confounders—other factors that are influencing the result.
A weight-loss pill company might boast that its users lost an average of 15 pounds in three months. The ad shows a convincing graph and glowing testimonials. What they intentionally don’t mention are the confounders. The people motivated enough to buy diet pills were probably also motivated to start eating healthier and exercising more. The company takes full credit for the weight loss, creating a causal link to their pill by simply hiding the other, more likely causes. They aren’t lying about the result, but they are masterfully faking its cause.
Ultimately, faking causality is a powerful form of storytelling. By cherry-picking the hero and hiding the messy details, anyone with an agenda—from a political strategist to a diet pill company—can craft a convincing narrative. These aren’t just statistical mistakes; they are carefully constructed illusions designed to sell a product, win a promotion, or claim credit that isn’t due. Recognizing the script is the first step to not falling for the performance.
]]>在中国的本科教育中，有一门很常见但形式差异巨大的暑期课程，叫做《社会实践》。不同学校、不同院系对它的理解和组织方式差异非常大。以我当时本科的北京大学经济学院为例，这门课的形式包括：老师联系地方企业或政府部门，组织学生实地参访；也可以由学生自主联系社会资源，自主选题开展调研。最终的考核往往是一份调研报告，通过就能获得课程学分。
类似的形式在美国也有，比如一些NGO与大学合作，邀请本科生参与社区议题的暑期项目。我本科时也带过两次社会实践队伍，体验非常丰富，也留下不少反思。
前不久旅行途中，在旅游大巴上就在琢磨：如果我来带《社会实践》，我会怎么设计，让学生能尽可能多地获得尝试、体验和学习？反过来说，也是在反思我自己当年做队长时，有哪些地方可以做得更好。
我觉得关键在两点：问题意识，以及理论与现实的对接能力。尤其对于经济学学生来说，这两点对他们理解理论、也理解社会运行方式都至关重要。
以“教育”这个方向为例，如果有学生对它感兴趣，那我们可以引导他们结合实践地点进一步细化研究问题。比如：
“教育真的能改变命运吗？”
“在农村，教育机会在性别之间存在差异吗？这种差异是如何产生的？”
这些问题都来自学生的直觉，但可以通过讨论引导，变成清晰的、有研究价值的调研目标。重点是让学生围绕，聚焦一个小问题，花整个夏天在社会实践中认真思考、认真回答。
当学生明确一个“接地气“的现实问题之后，老师可以进一步鼓励他们将问题抽象出来，对接已有的经济学框架。比如“教育能否改变命运”这个问题，其实可以纳入“人力资本投资”理论来讨论。
在调研之前，可以引导让学生先钻进理论里：阅读相关文献，理解这个理论模型是怎么设定假设、得出结论的。然后把这个理论作为工具，去指导调研设计、访谈提纲、数据采集等。但我认为更重要的一步，是最后要鼓励学生回过头来批判性地思考这个理论：它漏掉了什么？它和现实差距有多大？为什么差距会存在？这个过程，才是理论和实践真正的碰撞与融合。
每日讨论：建议每天调研结束后，大家围坐一圈，每人分享一点今天的所见所思。这种即时的分享可以带出不同视角，也帮助形成团队内的思维碰撞。
每日随笔：鼓励学生每天写一段短小的反思随笔，哪怕只有两三百字，最后放进实践报告里面去。写作是整理思路、锤炼逻辑的过程，也会在无形中留下他们思考的痕迹。
社会实践是一门可以极其丰富、也可以极其敷衍的课。关键不在于去的地方多么“高大上”，也不在于做出了多少“成果”，而在于学生有没有带着问题出发，带着新的视角返回。希望更多的《社会实践》课程，能够真正帮助学生站到现实中去，带着理论一起思考这个世界。
]]>王朝覆灭的最核心引擎，深植于古代社会的经济基础与专制政府“攫取的手”之间的矛盾。中国古代农耕社会的人均生产力水平极其有限，农民辛苦劳作一年的产出，往往仅够勉强维持家庭温饱，甚至略有不足。农业帝国的财政汲取能力，天然地被框定在一个极其狭窄的“生存阈值”之内。一旦统治者的需求超越了这个阈值，整个王朝系统便会陷入崩溃。
帝国战争：财政绞肉机： 帝国庞大的身躯运转需要巨大的财政支撑——维持常备军、供养官僚体系、进行大型公共工程（如长城、大运河）、以及应对突发灾难或外敌入侵。这些开支如同沉重的负担，最终都转嫁到以农民为主体的纳税人肩上。这里面最“人祸”的往往是帝国扩张的欲望，或者说边疆少数民族的侵扰。无论是汉武帝雄心勃勃的北击匈奴、唐玄宗晚年的怛罗斯之战与安史之乱平叛，还是明朝万历年间耗费巨大的“三大征”（宁夏之役、播州之役、朝鲜之役），抑或是为防御游牧民族而被动修筑长城、漕运系统，战争或大规模国防工程都是吞噬国帑的无底洞。汉武帝晚年因连年征战导致“海内虚耗，户口减半”，国家财政濒临崩溃。著名的《轮台罪己诏》，调整政策以求喘息。北宋为了抵御辽、西夏，维持了庞大的常备军（禁军、厢军总数常达百万以上）形成了“冗兵”的沉重财政负担和器械费用。而战争机器一旦发动，除非战争胜利，否则或迫于当时的政治形势，或迫于君王的“身后英明”，这个“吞金兽”都很难停下来。
恶性循环：赋税、起义与崩溃： 巨大的财政缺口迫使统治者加税、摊派、预征（如明朝末年的“三饷”——辽饷、剿饷、练饷）。当赋税超过了农民维持生存的底线，流离失所、卖儿鬻女，人肉相食便成为常态。最终，走投无路的农民/土匪揭竿而起（如秦末陈胜吴广、唐末黄巢、元末红巾军、明末李自成张献忠）。大规模的起义需要更多的军队去镇压，这又需要更多的军费，只能再次加税，进一步摧毁生产力和税基，形成“财政危机→加税征役→民变起义→负担继续加重”的的恶性循环。帝国有限的军事资源在镇压内乱和抵御外侮之间疲于奔命，最终两头落空，摧毁了原本稳固的政权，引发朝代更替。崇祯皇帝在内外交困中苦苦支撑，加征的“三饷”成为压垮大明王朝的重要稻草，正是这一循环的经典写照。
专制国家既要用人更要防人，统治者在提拔名将与防范武人之间左右为难。如何掌控这把锋利的双刃剑，成了专制皇权永恒的难题——一个困扰古今的“委托-代理”困境。倘若信任武将强化战力，则可能滋长割据之心；若轻视武将，则军力薄弱难以御敌。
放权与失控：名将的双刃剑： 一方面，为应对大规模战争或边境危机，皇帝必须赋予将领足够的指挥权、财权和人事权（如汉末的州牧、唐中后期的节度使）。然而，权力一旦下放，极易滋生拥兵自重、尾大不掉的局面。唐玄宗开元年间设立藩镇节度使以戍边，但节度使长期掌握地方兵权和财政税收，史书称他们“既有其土地，又有其人民，又有其甲兵，又有其财赋”，因而成为威胁中央的隐患。到了安史之乱后，藩镇势力更加坐大，军阀的扩张一度让中央难以控制。明朝的燕王朱棣，凭借自身强大的军事实力，通过“靖难之役”从侄子建文帝手中夺取了皇位。明末则出现了皇帝猜忌将领的案例：崇祯帝因多疑处死了镇守辽东的名将袁崇焕，导致边防崩溃。
集权与弱化：自废武功的无奈： 深谙历史教训的开国君主们，往往在政权稳固后着手削弱、收夺将领兵权，以防陈桥兵变重演。宋太祖赵匡胤“杯酒释兵权”，以优厚待遇换取开国元勋的兵权，并确立“以文制武”、“更戍法”等制度，有效防止了武将专权，但也导致了军队战斗力下降、指挥效率低下，面对辽、金、西夏的军事压力时屡屡处于劣势。明太祖朱元璋和明成祖朱棣通过“蓝玉案”等残酷手段清洗功臣宿将，将军事指挥权高度集中于皇帝和兵部，后期依靠文臣督师（如孙承宗、袁崇焕），同样面临将领临机决断权不足、军队活力下降的问题。文官指挥虽加强了皇权控制，却难以匹敌如努尔哈赤、皇太极这样在战争中成长起来的军事天才及其高效的组织。
现代回响：中东的军政府困境： 这一难题并非古代中国独有。当代中东许多国家（如埃及、叙利亚、伊拉克在特定历史时期）的军政府，同样面临如何在维持军队战斗力的同时，防止军人干政甚至颠覆文官政府的挑战。纳赛尔、萨达姆等强人依赖军队上台并维持统治，其权力基础与军事力量深度绑定，但也始终无法摆脱被军队内部力量挑战的风险。
皇权高度集中同时也埋下了许多结构性隐患。专制制度将最高权力集中于皇帝一人。如何平稳实现最高权力的交接，成为帝国稳定的核心问题。然而，围绕皇位继承的争斗，几乎伴随着每一个王朝的始终。
继承人难题：国本之争： 皇帝子嗣众多、嫡庶有别、长幼有序的规则常被打破。秦始皇突然病逝，遗诏被篡改，引发胡亥、扶苏之争，加速了秦朝崩溃。汉武帝的”废太子案“，先铲除了一批”太子党“，再铲除一波”反太子党”，导致朝中无人可用。康熙晚年的“九子夺嫡”更是惊心动魄，虽以雍正继位告终，但消耗了巨大的政治资源。
外戚与权臣：皇权的延伸与挑战： 一旦皇帝驾崩或年幼继位，朝廷就容易出现摄政与外戚干政。皇帝依赖近亲（母族、妻族）或亲信大臣辅政，本意是巩固皇权。但这些人极易利用接近权力核心的优势，发展成庞大的政治集团，甚至架空皇权。西汉末年的王莽以外戚身份（汉元帝皇后王政君之侄）逐渐掌握大权，最终篡汉自立。东汉中后期，外戚（如窦宪、梁冀）与宦官交替专权，朝政混乱不堪。总之，这些内部纷争使统治效率下降，常常动摇政局，但归根结底它们都发生在专制体制框架内，有导致朝代更替的案例，但并不是很多。
在分析王朝衰亡时，常有人将矛头指向文官集团（相权）与皇权的矛盾，认为这是导致分裂或衰弱的因素。然而，皇帝与相臣的权力斗争、文臣集团的兴衰更迭，本质上也都服务于专制体制本身。
相权的本质：皇权的执行官： 丞相或内阁首辅的权力，从根本上源于皇帝的授权。其职责是协助皇帝处理庞大帝国的日常政务，是专制机器得以运转的核心部件。历史上权相辈出（如李林甫、蔡京、严嵩、张居正），归根到底也是服务于专制制度，从未发展出脱离皇权、分庭抗礼甚至取而代之的制度性诉求。严嵩与徐阶、高拱与张居正的权力更迭，本质上是皇帝驾驭下，不同官僚集团或个人在专制体系内部的兴衰沉浮，是“代理人”之间的竞争，而非相权作为一种制度力量对皇权的根本性挑战。
儒学的粘合剂：君臣纲常的驯化： 儒家思想，特别是其核心的“三纲五常”伦理，为这种服务于专制制度的君臣关系提供了强大的意识形态支撑。“君为臣纲”如同铁律，深深植入士大夫的精神世界。忠君即是爱国，维护皇权统治的稳定是士大夫阶层天然的道德责任和政治使命。科举制度更进一步将天下英才纳入这一体系，通过功名利禄的激励和儒家伦理的教化，使整个文官集团成为专制制度最忠诚、最有效的统治工具。
当我们回望历史周期律，生产力的束缚与帝国财税攫取的手是深层地基，军事控制的悖论与权力顶端的脆弱则如同地基上的裂痕，共同决定了王朝大厦的高度与最终倾颓的时刻。那么，在“秦制”框架已然存在的前提下，能否尝试破解这看似宿命的周期律？若将“专制”视为一个难以撼动的前提（如何超越专制本身，则是另一个更为宏大的命题），历史的经验或许提示了三条可能的缓解路径：
克制欲望，涵养民力： 效法汉文帝、景帝的“文景之治”，奉行“休养生息，无为而治”的国策。关键在于专制帝王能否主动克制其“帝国扩张欲”，避免无休止的征伐与宏大工程，将国家财政和民力消耗控制在可持续的范围内，让社会得以喘息和积累。
强化控制，驾驭武力： 借鉴清朝“军机处-密折制度” 的经验，构建严密的信息控制与垂直指挥系统。虽然统领军队的仍需专业武官，但通过制度设计（如频繁调动、分权制衡、直达天听的密报网络），确保皇帝能近乎实时地掌握所有关键信息，并牢牢控制着将领的任免与军队的动向，最大程度抑制军阀化倾向。
规范传承，减少内耗： 建立更完善、更透明的皇位继承制度。在无法摆脱“家天下”逻辑的前提下，通过明确规则（如嫡长子继承制的严格执行，或类似清朝“秘密立储”的创新尝试）和早期培养，尽可能减少围绕最高权力交接的不确定性和血腥内斗，避免因继承人危机或宫廷政变导致的政治资源巨大浪费与统治核心的分裂。
这些策略，本质上都是在专制制度的“天花板”下，试图修补其最易溃决的堤坝。它们或许能延长王朝的“健康期”，缓和某些危机的烈度，甚至在特定条件下缔造如“康乾盛世”般的辉煌。然而，它们更像是精妙的“保守疗法”，在延缓衰败的同时，也可能强化了系统的僵化。当遭遇前所未有的外部冲击（如工业文明）或内部积弊突破临界点时，周期律的巨轮依然会无情碾过。
（Thanks to Chatgpt and Deepseek)
]]>Recently, I was chatting with a friend when driving a car, while listening the podcast together at the same time. Must be inspired by some topics in the bodcast, my friend shared a “dark fact” by his friend. He told me that, his friend’s family is doing international trade, particularly about shipping. So his friend told him that, strictly speaking, almost all Chinese foreign-trade firms are now doing unlawful tariff avoidance under the current law and regulations. But the government just turns a blind eye. I have no clue whether this is the fact or being exaggerated, but I will not be surprised if it is true. It makes perfect sense from an economic perspective!
Most economics model/theory are math formalizations of human beings intuitions. What I want to say follow, From the formal it is called two-stage dynamic games. But just plainly speaking, the legistration department has incentive to make the law/regulation DETAILED, STRICT and UNREASONABLE at the first place. This is because with this law, the government has the room to play strategically in the second-stage: if the government feels beneficial or necessary, whether from “good incentive” (e.g. to promote market fairness) or “bad incentive” (e.g. to fine firms and increase government income, “the extractive hand”), it can enforce the law “strictly”; if the government feels not beneficial to strict the law (e.g. to protect these international trade firms), it can just turn their eyes blind. Moreover, this creates the room for rent-seeking, or heterogeneous treatment : for some firms (e.g. with connection, obedient or bribary) the government can choose to not enforce the law; while it can enforce the law and punish other firms. In addition, this can even let the government catch the dirt of firms. The government has the opportnity to punish firms
The key, as well as the necessary condition in this story is the legistration and enforcement are somewhat the same party (i.e. the party or the government) in the story, or at least they share a large proportion of the same interest. For example, in U.S. the legislation (the house and senate) and the enforcement (police) are diffrerent and share no common interest. This ensures that the legislation will not be too strict.
]]>