1.

Quantative Methods
R1: Multiple Regressions

Bacis and Underlying Assumptions

Evaluating and Interpret Regression Models
Model Specification
Extensions Of Multiple Regression
R2: Time-Series Analysis

Linear and Log-linear Trend Models

AutoRegressive(AR) models
R3: Machine Learning
R4: Big Data Projects
Comments
2, Economics
R5: Currency Exchange Rate
Cross Rate
Mark-to-market Value and Parity Conditions
Carry Trade, BOP, Exchange Rate Derminants
R6: Economic Growth
Growth Factors and Production Function
Growth Accounting and Influencing Factors
Growth and Convergence Theories
R7: Economics of Regulation
Comments

1. Quantative Methods
Normal Distribution

正态分布中间概率confidence interval对应的标准差的倍数值68% (1.0)、90%(1.65)、95%(1.96)、98%(2.33)%，

99%(2.58).

Skewness 指尾部那边长，左偏or负偏就是指左边低长、右边高。

Skewness is the third central moment in a return distribution that is often used to measure the
distribution’s departure from symmetry
A distribution can have right (or positive), left (or negative), or zero skewness. A right-skewed distribution is
longer on the right side of its peak, and a left-skewed distribution is longer on the left side of its peak。
 上图从左到右依次是：positive、zero、negative skewness。
In a negatively skewed distribution: Mode众数 > Median中位数 > mean平均数

Trend following and the diversification potential of positive skew

https://www.caisgroup.com/articles/trend-following-and-the-diversification-potential-of-positive-skew

R1: Multiple Regressions

Bacis and Underlying Assumptions

Basic concepts

Dependent variable (Y,因变量), also know as explained variable, predicted variable

Independent variable(X,自变量), also known as explanatory variable, predicting variable.

Assumptions underlying a multiple regression model include:

A linear relationship exists between the dependent and independent variables.

The residuals are normally distributed.
The variance of the error terms is constant for all observations.
The residual for one observation is not correlated with that of another observation.
The independent variables are not random, and there is no exact linear relation between any two or more
independent variables.

Measures of goodness of fit

RSS(Regression of Sum Squared Variations) + SSE(Sum of Squared Errors) = SST(Sum of squared Total varisions)
SSE = ∑(yi − y^i )2 . This gives a measure of the total deviation of the predictions from the actual values.
​ ​ ​

SSE also known as the Sum of Squared Residuals (SSR) or the redisual sum of squares。
RSS = ∑(y^i − Yˉ )2 measure the variance in the dependent variable that is explained by the independent
​ ​

variable x.
SST = ∑(yi − Yˉ )2 measure the total variance in the dependent variable y.
​

yi actural value, y^i predicted value, Yˉ

​ ​ ​

Mean Squared Regression: MSR = RSS/k , k=1 for single lienar regression
Mean Squared Errors： MSE = SSE/(n − k − 1) ,
 Standard Error of Estimate/Regression
SSE ∑ni=1 (yi − y^)2
SEE = MSE = =
​ ​ ​

n−k−1 ​ ​

n−k−1 ​ ​

SEE : the standard deviation of error terms in the regression.

The standard error of the regression is also known as residual standard error. 说的就是SEE。
计算公式：所有样本点预测值与真实值的残差序列，平方后除以n-k-1，k是自变量的个数，然后再开方
SEE measure the degree of variablity of the actual Y-values relative to the estimated Y-values from a
regression equation.
Roughly 95% of the observation should fall within +/- two standard error of the regression, which is a
quick approximation of a 95% prediction interval.
如果所有预测值y变为原先的50%，SEE（the standard Error of the regression)也为原先的50%，但是R2不变
The smaller the SEE, the better the fit.

Evaluating and Interpret Regression Models

Analysis of variance (ANOVA) is a statistial procedure that provides this information.

ANOVA Table
n: number of observations
k: number of independent variables
Standard Error 就是Standard Error of Estimate/Regression
The standard error of regression coefficients： SE(β^j ) ,tell us how much sampling variation there is if we
​ ​

were to re-sample and re-estimate βj 。就是从总样本集中多次采样拟合，得到一系列预测的 βj ，然后这个序

​ ​

列的Standard Error。最终用的 βj 是一系列的均值。 ​

SE(β^j ) 可以推导出不同置信概率下的取值范围。
​ ​

βj −0
The t-statistic for a slope coefficient βj : =
​

SEβj ​
​

就是在做假设检验时，若空假设null hypothesis是这个系数=0，那么对应的t-statistic是多少。
如果一个系数的t-statistic的绝对值 > 1.0， 那么adjusted-R2 增加；反之这个系数的t-statistic的绝对值 <
1.0， 那么adjusted-R2减少。
 The p-value for a slop coefficient: p-value = P(t>t.Stat_{beta_j})，有时等于P/2注意One-tail或者Two-tail
p-value越低，代表空假设发生的概率越低，p-value <significant level alpha, 则可以拒绝空假设。

While the adjusted R2 penalizes overfiting, it does not indicate the quality of model fit(like AIC/BIC), nor does it
indicate statistical significance of the slope coefficients. We can formally evaluate the overall model fit using an F-
test (discussed later).

R2 Coefficient of Determination

R2 = Explained variation
T otal variation ​ = RSS
SST ​ = SST −SSE
SST ​ = T otal Variation−U nexplained variation
T otal variation ​

评估所有自变量对因变量的整体解释程度。值越大，模型拟合的越好

Adjusted Ra2 adjusted for degrees of freedom, penalizes overfitting

​

Adjusted R2 (Ra2 ) = 1 − [( n−k−1

​
n−1
⋅ (1 − R2 )] ​

Ra2 <= R2 。增加模型independent variable总是会增加R2，但 Ra2 可能会增加或降低

​ ​

Both AIC and BIC evaluate the quality of model fit among competing models for the same dependent variable.

Akaike’s informaon criterion (AIC) is used if the goal is to have a better forecast, while the Schwarz’s
Bayesian information criteria (BIC) is used if the goal is a better goodness of fit.
AIC = n ln( SSE
n ) + 2(k + 1) ​

BIC = n ln( SSE

n ) + ln(n) (k + 1)​

BIC对变量个数的惩罚因子更重。
AIC 和BIC都是值越小、模型越好

F-test（is a joint hypothesis test） 评估多个变量共同的作用，不像t-test那样评估单个independent variable的影响。

the Unrestricted model（or the full mdoel， nested model）：原始含有多个k个自变量模型。

The Restricted model：只含有一部分自变量的模型，排除q个 independent variables.

(SSER −SSEU )/q

F-statistic =
​ ​

SSEU /(n−k−1)
​
​

R and U represent the restricted and unrestricted models, repectively

 q = number of excluded variables in the restricted model
k = total independent variables in the unrestricted model
An F-statistic评估这q个自变量整体对因变量y的解释程度。经常被用测试q个自变量中是否至少有一个自变量对因变
量y有显著的解释或影响。An F-statistic asseses how the independent variables, as a group, explains the
variation in the dependent variable; or used to test whether at least one independent variable explains a
significant portion of the variation of the dependent variable.
注意q=k，即所有自变量都排除 SSER = SST （all of the variation in the restricted model is unexplained），
​

测模型是否至少有一个变量得系数显著不等于零。

Linear F-test（原假设一般为几个变量的系数=0，如b2=b3=0),满足下列条件拒绝原假设。

F-statistic > F-critical value(One tail),拒绝原假设。 表明排除的q个变量中至少一个变量的系数slop coefficient显

著不等于零（is statistically different zero）
p-value < significant value( α ), 拒绝原假设，for t-test
注意查找Partial F-table(5% level of signifcance)表中分子numberator degrees of freedom = q，分母
dominator degrees of freedom = n-k-1对应的值为对应F-critical value。

Model Specification

Regression model specification is the selection of the explanatory (independent) variables to be included in a
model, and the transformations (if any) of those explanatory variables.

Different types of functional form misspecifications.

Omitting a independent Variable

Variable Should Be Transformed
Inappropriate Scaling of the Variable
Incorrectly Pooling Data
指Sample has periods of dissimilar economic environments(that should not be pooled.)

There are three primary assumption violations that you will encounter:

(1) heteroskedasticity,
(2) serial correlation (i.e., autocorrelation),
(3) multicollinearity.
For ecach assumption answer the following qustions：

a. What is it?

b. What is its effect on regression analysis?

c. How do we detect it?

 d. How do we correct for it?

Heteroskedasticity 异方差性，就是不满足残差的同质性。 the variance of the residuals is not the same across all
observations in the sample

Unconditional heteroskedasticity ：无条件的异方差性 occurs when the heteroskedasticity is not realted to

the level of independent variables。
While this is a violation of the equal variance assumption, it usually causes no major problems with the
regression.
Conditional heteroskedasticity：异方差性realted to the level of independent variables
例如异方差性随着自变量X值的变大呈变大趋势。 It does create significant problems for statistical
inference.

Effect of Heteroskedasticity on Regression Analysis

The standard errors are usually unreliable estimates. (For financial data, these standard errors are usually
underestimated, resulting in Type I errors.)
The F-test for the overall model is also unreliable.

Two methods to detect heteroskedasticity:

by examining scatter plots of the residuals

Breusch–Pagan (BP) chi-square (χ2) test to detect conditional heteroskedasticity. 核心思想是用残差项的平方
来替代原来的residual，重新训练
BP chi-square test statistic = n * Rresid
2 with k degrees of freedom
​

2
Rresid
​ =R2 from a second regression (of the squared residuals from the first regression) on the
independent variables
 BP test statistic 大于 对应的critical value，才拒绝原假设（没有condicitonal heteroskedasticity），结论原
数据有condicitonal heteroskedasticity。

Autocorrelation

Serial correlation, also known as autocorrelation, refers to a situation in which regression residual terms are
correlated with one another; that is, not independent. Serial correlation can pose serious problem with
regressions using time series data.

Detecting Serial Correlation

Residual serial correlation at a single lag can be detected using the Durbin–Watson (DW) statistic. A more
general test (which can accommodate serial correlation at multiple lags) is the Breusch–Godfrey (BG) test.
The DW statistic is designed to detect positive serial correlation of the errors of a regression equation.
DW statistic < critical value,则存在serial correlation
The Breusch–Godfrey (BG) test is for serial correlation.
the BG test statistic > the critical value, 则存在serial correlation。
In the presence of serial correlation, if the independent variable is a lagged value of the dependent
variable, then regression coefficient estimates are invalid and coefficients’ standard errors are deflated, so
t-statistics are inflated.

Durbin Waston Test用于回归模型中的自回归监测；看AR(p)模型是否选对

For instance, when lagged dependent variables are included in the explanatory variables, then it is
inappropriate to use this test.(Durbin Waston Test)
计算方式：给定一个回归模型，n个序列样本；①计算预测误差的平方和SSE， ②然后每个样本预测误差（y‘ -
y_true）减去前面那个样本的误差、得到difference，计算这n-1个difference平方的和。最后②除以①，就是durbin
Waston的值。
DW statistic = 2.0， indicating zero autocorrelation； > 2.0 indicates negative autocorrelation; below 2.0
mean there is positive autocorrelation
https://www.investopedia.com/terms/d/durbin-watson-statistic.asp#:~:text=The Durbin Watson statistic
is,above 2.0 indicates negative autocorrelation.

Multicollinearity 共线性

Multicollinearity值多个自变量之间的共线性
Detecting Multicollinearity
VIF(variance inflation factor) 排除自变量j重新算剩余变量的R2
V IFj = 1/(1 − Rj2 ), Rj2 值越大（共线约严重）、 V IFj 值越大
​ ​ ​ ​

V IFj > 10 indicates serious multicollinearity issues requiring correction; V IFj > 5 warrants further
​ ​

investigation of the given variable.

Extensions Of Multiple Regression

Influence analysis and methods of detecting influential data points.

异常点Y值都是有一个或一些异常值X_j引起的。

Leverage is a measure of the distance between the j th observaton of independent variable i relative to its
sample mean.
 The sum of the individual leverages for all observations is k + 1. If a variable’s leverage is higher than
three times the average, [3(k + 1) / n], it is considered potentially influential.

Indentify outliers using the studentized residuals

any observation with a studentized residual whose absolute value exceeds the critical t-value is a
potentially influential observation.

Cook’s distance (Di) is a composite metric (i.e., it takes into account both the leverage and outliers) for
evaluating if a specific observation is influential.

e2i
Di =
​
​

​
hii
k⋅MSE [ (1−h2ii ) ] ​
​ ，一般不要求计算，会直接给出

Di > 2 ∗
​ k/n indicate that the ith observation is highly likely to be an influential data point. 注意这里的系
​

数2 是乘以。
k = independent variables的个数
n = 样本点个数 (#observations)

R2: Time-Series Analysis

Cross-section data vs. time-series data

Linear and Log-linear Trend Models

Linear and Log-linear Trend Models

Linear trend model： yt = b0 + b1 ⋅ t + ϵt
​ ​ ​ ​

where, t= time(independent variable), t= 1,2,3...

Log-linear trend model： ln(yt ) = b0 + b1 ⋅ t + ϵt ​ ​ ​ ​

where, t=1,2,3...
Choose： When a variable grows at a constant rate, a log-linear model is most appropriate. When the
variable increases over time by a constant amount, a linear trend model is most appropriate.
Limitations：
the assumptions underlying linear regression is that the residuals are uncorrelated with each other.
但当样本的残差之间是autocorrelation自相关时，不能用这个模型。

AutoRegressive(AR) models

When the dependent variable is regressed against one or more lagged values of itself, the resultant model is called
as an autoregressive model (AR).
用AR model的前提是time series being modeled is covariance stationary.

Covariance stationary 协方差平稳

定义：A time series is covariance stationary if it satisfies the following conditions:

序列的均值和方差是常数，不随着时间t改变而改变。Constant and finite expected value and finite variacne.
A time series must have a finite mean-reverting level to be covariance stationary.
序列中任何两个term的协方差 COV (xi , xi+k ) = γ(k), f or ∀i ∈ 1, 2, 3... 只与相对位置j有关、与i无关。即
​ ​

COV (x1 , x1+k ) = COV (x2 , x2+k ) 。The covariance between any two terms of the sequence depends only
​ ​ ​ ​

on the relative position of the two terms and not on their absolute position. Constant and finite
covariance between values at any given lag.
 备注：协方差是两个随机变量之间线性相关性的度量。它比较两个随机变量与其平均值（或预期）值的偏差。随
机变量X和Y的协方差公式： COV (X, Y ) = E(X − μX )E(Y − μY ) ​ ​

AutoRegressinve(AR) model - > ARIMA

AR(1) model xt = b0 + b1 xt−1 + ϵt

​ ​ ​ ​ ​

AR(p) model xt = b0 + b1 xt−1 + b2 xt−2 + ... + bp xt−p + ϵt

​ ​ ​ ​ ​ ​ ​ ​ ​

p indicates the number of lagged values that the autoregressive model will include as independent
variables.
Forecasting with AR model:地推式逐个计算，namely chain rule of forecasting.
Autocorrelation & Model Fit
When an AR model is correctly specified, the residual terms will not exhibit serial correlation
Test whether an AR model is correctly specified：每个样本点对应的residuals、计算lag1,2,3的
autocorrelation，然后test whether the autocorrelations are significantly different from zero:t-test看否显
著（也可画图看分布）
其中t-statistic is the estimated autocorrelation divided by the standard error. The standard error is
1/ n , n 就是观察点数据个数。不同lag 的standard Error一样，根据standard error可以求观察点个数。
​

Financial and economic time series inherently exhibit some form of instability or nonstationarity.金融环
境是动态变化的、不同时期模型的系数都不一样。

Random Walk and Unit Roots

If a time series follows a random walk process (with a drift), the predicted value is equal to the value in the
previous period plus a random error term.
Random Walk xt = b0 + b1 xt−1 + ϵt
​ ​ ​ ​ ​

b0 the constant drift, =0 表示without dirft

​

b1 = 1 , The time series is said to be have a unit root and will follow a random walk process.
​

a random walk (with/out a drift) is not covariance stationary. 因为the mean-reverting level is b0​

1−b1 ​
​ , 不是
常数 (the division of any number by zero is undefined)

To determine whether a time series is covariance stationary, we can

(1) run an AR model and examine autocorrelations,

A stationary process will usually have residual autocorrelations insignificantly different from zero at all
lags or residual autocorrelations that decay to zero as the number of lags increases.
(2) perform the Dickey Fuller test. （Unit Root Testing for Nonstationarity）
Dickey Fuller test. 将测试 b1 = 1 转化为 b1 − 1 = 0 的null hypothesis. ​ ​

Covariance stationarity and Seasonality

Covariance stationarity 协方差平稳

一般的AR model适用的都不是协方差平稳的吧，但它的一阶差分first differencing、二阶差分一般就是

covariance stationary.
Covariance stationarity can often be achieved by transforming the data using first differencing and
modeling the first-differenced time series as an autoregressive time series.
When a covariance-stationary series is at its mean-reverting level, the series will tend not to change
until it receives a shock. In an AR(1) model, the general expression for the mean-reverting level is b0/(1
− b1).
 该稳定的时间序列随时时间推移、最终会达到它均衡的值mean-reverting level。 A stationary time series
may need many periods to return to its equilibrium, mean-reverting level.

Seasonality in a time-series is a pattern that tends to repeat from year to year.

Detecting seasonality: 观察residual lag 1,2...大小

Correcting for seasonality. 添加对应的x_{t-k}项

ARCH and Multiple Time Series

Autoregressive conditional heteroskedasticity (ARCH)

如果残差的方差依赖其前一个值，就用这个模型。 if the variance of the residuals in one period is
dependent on the variance of the residuals in a previous period.
ϵ2t = a0 + a1 ϵ2t−1 + μt
​ ​ ​ ​ ​

Cointegration means that two time series are economically linked (related to the same macro variables) or
follow the same trend and that relationship is not expected to change

R3: Machine Learning

Types of Learning and Overfitting Problems

Inconsistency error 数据不一致、有误

An inconsistency error occurs when a data point conflicts with corresponding data points or reality.

Supervised Learning Algorithm

Penalized regressions:正则化防止过拟合
Least absolute shrinkage and selection operator (LASSO).
SVM， KNN，CART...
Ensemble and Random Forest

Unsupervised Learning Algorithms

Principal component analysis (PCA)

Clustering
K-means clustering
Hierarchical clustering

Other Models

Neural Networks
Deep Learning Networks (DLNs)
Reinforcement Learning (RL)
TF-IDF（ term frequency–inverse document frequency）
Term Frequency TF (t, d) = ∑ ′ft,df ′ , is the relative frequency of term t within document d。就是term在文档
​

t ∈d ​

t ,d ​

中出现的次数，不是在数据集中出现的次数。分母是所有term在文档d中出现的次数之和。
注意TF calculation is at the document(sentence) level, not at collection level(all documents).
注意TF分很高的token往往是stop/common words, TF分很低的token往往是sparse terms（专有名词、地名
等）。TF分中间部分的tokens areimportant to the meaning of the text
 Inverse Document Frequency IDF (t, D) = log N
∣{d∈D:t∈d}∣ ​ 。就是term t出现在多个文档中除以文档的总个数N，然
后再求倒数inverse，最后取log平滑。

R4: Big Data Projects

Big data is characterized by the three Vs of volume, variety, and velocity:

Data Analysis Steps / Build Model Steps

1、Conceptualization of the modeling task.
2、Data(or Text) collection.
Data (text) curation: 从网上或其他渠道爬取或收集原始信息。 The process of gathering relevant external
text data via web services or web-spidering (scraping or crawling) programs that extract raw content
from a source。
3、Data(or Text) preparation and wrangling(processing).
Text preparation(cleansing): remove HTML tags, punctuations,white blanks. Numbers should be
removed or substituted with an annotation /number/.
Text warngling(processing): Lowercasing, Remove stop words, Stemming, Lemmatization(lemma)
data transformation and scaling.
4、Data(or Text) exploration.
feature selection and engineering
5、Model training.
These above five steps are iterative.
Text problem formulation is the process of determining how to formulate the text classification
problem and identifying the exact inputs and outputs for the model.
Data Exploration
1、Exploratory data analysis (EDA)
2、Feature selection：
Frequency, Chi
Chi-square
Mutual infomation（MI）: If the token appears in all classes, it is not considered a useful discriminant,
and its MI equals 0. Tokens associated with only one or a few classes would have MI approaching 1.
3、Feature engineering: create new features by transforming, decomposing or combining multiple
features.
Model Training and Evaluation
1、mothod/model selection
2、Performance valuation
Error analysis: confusion matrix, precision(P), recall(R), F 1 score = 2/( P1 +
​
1
R)
​ = 2∗P ∗R
P +R ​

Receiver operating characteristic (ROC).

Root mean square error (RMSE).
....
3、Tuning
Hyperparameters, grid search

Comments
The data (text) curation step involves gathering relevant external text data via web services or programs that
extract raw content from a source.
 the text preparation and wrangling step
involves cleansing, preprocessing, and converting the data into a structured format usable for model
training.
remove numbers, perform Stemming and lemmatization.

2, Economics

R5: Currency Exchange Rate

Cross Rate

Bid-Ask Spread： If the quote in the interbank USD/EUR spot market is 1.3649/1.3651

Bid-Ask Spread = ask price(1.3651) - bid price(1.3649) =0.0002(two “pips” wide)

前面1.3649表示银行bid买入价（就是顾客的卖出价）,后者1.3651是银行ask/offer price卖出家（就是顾客的买入价）
银行作为dealer一般是买低卖高赚取差价

银行作为dealer对个人的市场， 银行间的市场interbank exchange market。

交叉汇率相关
Cross rates
Given: RMB/USD = 7.0, USD/AUD = 0.6
Get:
RMB/AUD = RMB
U SD ​ ∗ U SD
AU D ​ = 7.0 ∗ 0.6 = 4.2
1 1
AUD/RMB = U SD
RMB ​ ∗ AU D
U SD ​ = 7.0 ​ ∗ 0.6 ​ = 1/4.2 = 0.238
Cross rates With Bid-Ask Spreads计算口诀：相乘同边、相除对角，乘小除大。
Rule1： ( CA)
bid = ( B )bid ⋅ ( C )bid , ( C )ask = ( B )ask ⋅ ( C )ask
​ ​
A ​
B​ ​
A ​
A B ​ ​ ​ ​ ​ ​

1 1
Rule2: 同汇率bid-ask互换， ( B
C )bid =
C
(B )ask
， (B
C )ask = ( B
C
)bid
​ ​

​ ​
​ ​ ​

​ ​
​

理解：对于银行，商品C货币的买价 ( C )bid 等于商品B货币的卖价 ( C 1)

B
​ ​ ​ 的倒数。
B ask ​ ​
 Triangular arbitrage
基于A/B,B/C得到(A/C)'的bid-ask spread 和 直接A/C的 bid-ask区间 不重叠，则有套利机会。
如上例计算方法是以B货币出发，若(A/C)' < A/C,表示A/B,B/C市场中C货币相对低，（买低卖高），应该用B先
基于B/C 兑换低的C，在贵的市场A/C兑换A（相当于卖C），最后基于A/B兑换为新的B’，B'-B即为利率。
就是B货币在一个市场买价格低的货币，再另外一个市场卖出，最后再换汇回来。

Mark-to-market Value and Parity Conditions

forward premium (discount) = F − S0

[F Pt (T )−F P0 (T )] (contract size)

Mark-to-Market Value Vt =
​ ​

1+R( days
360 )
​ ​

Vt 就是先算forward contract到期是T的价值，然后再折现到现在时间点t的价值。
​

F P0 (T ) 合约最初0时刻签订时，在T时刻的执行价格（购买base currency的价格）
​

F Pt (T ) 合约t时刻在T时刻的执行价格（卖出或卖出的价格）（若是汇率对应bid price, sell base currency)

​

days 当前时间t距离合约到期日T的天数
R = 年化annualized interest rate of price currency.
注意：
合约的价值是按price currency定的， base currency是“标的商品”。
这里FP对应最初签订合约价格的long方、即购买价格， F Pt 则用t时刻卖出价格，分子为 F Pt (T ) − F P0 (T ) ； ​ ​ ​

若果FP对应最初签订合约价格的short方、即卖出价格 F Pt 则需用t时刻买出价格（ask price），分子为 ​

F P0 (T ) − F Pt (T ) ； ​ ​

Interest Rate Parity (汇率-利率)

Covered Interest Rate Parity

Fx/y 1+rx
=
​
​

Sx/y ​
​

1+ry ​
​

The word ‘covered’ in the context of covered interest parity means bound by arbitrage. Convered指
受套利约束，uncovered指不受套利约束
公式中是名义利率Nominal interest rate
Uncovered Interest Rate Parity
E(St ) 1+rx
S0 ​
​

​ = 1+ry
​

​
​

哪种货币的利率高、未来就会相应的贬值。
If uncovered interest rate parity (and covered interest parity) holds, the forward rate is unbiased
predictor of future spot rate (i.e., forward rate parity holds)
One of the assumptions of uncovered interest rate parity is that investors are risk neutral.
Forward Rate Parity指 F = E(St ) ​

Forward rate = expected future spot rate (F=E(S1))

一般short term不满足，long-term有这个趋势
F −S0
S0 ​
​

​ = E(%ΔS)

Purchasing Power Parity(购买率平价公式： 汇率-通胀)

Absolute PPP：两国的汇率值等于两国CPI的比值
The law of one price states that identical goods should have the same price in all locations.同样的产品
在不同地区的价格应该是一样的！
SA/B = CP IA / CP IB , CPI一揽子产品的加权价格 ​ ​ ​
 Relative PPP：两国汇率的的变化等于两国通胀率变化的差。
%ΔSA/B ≈ Inf lationA − Inf lationB ,
​ ​ ​

%ΔSA/B = change in spot price(A/B)

​

%ΔSA/B = (Inf lationA − Inf lationB )/InfB ,

​ ​ ​ ​

A国通胀率相对高、则相对于B国货币贬值。两国今年通胀率一样、则汇率不变。
Relative PPP states that changes in exchange rates should exactly offset the price effects of any
inflation differential between two countries.

Fisher Effect (利率-通胀关系)

(Domestic) Fisher Relation

1 + Rnominal = (1 + Rreal )(1 + Einf lation ) ≈ Rreal + Einf lation
​ ​ ​ ​ ​

Rnominal ≈ Rreal + Einf lation

​ ​ ​

International Fisher Relation

Rnominal A − Rnominal B = Einf lation A − Einf lation B ,
​ ​ ​ ​

Rreal A ≈ Rreal B Under real interest rate parity, real interest rates are assumed to converge across
​ ​

different markets.

When Uncovered Interest Rate Parity and Purchasing Power Parity hold together, they illuminate a relationship
named real interest rate parity, which suggests that expected real interest rates represent expected adjustments
in the real exchange rate. This relationship generally holds strongly over longer terms and among emerging
market countries.

真实利率平价：汇率会随着真实利率而变动。

Carry Trade, BOP, Exchange Rate Derminants

Fx Carry Trade 外汇套息（利差）交易

利差交易指：假设当前A/B的spot rate 和 Forward rate一样，那么借利率低的货币、去投资利率高的货币

利差交易盈利的基础是Covered Interest Rate Parity。The carry trade is profitable only if uncovered interest
rate parity does not hold over the investment horizon.
the return distribution of the carry trade is not normal; it is characterized by negative skewness and excess
kurtosis (i.e., fat tails), meaning that the probability of a large loss is higher than the probability implied
under a normal distribution. We call this high probability of a large loss the crash risk of the carry trade.
In a negatively skewed distribution: Mode众数 > Median中位数 > mean平均数
A国利率相对高（货币有贬值倾向），但这会吸收更多资金流入：1）若引起经济发展、通胀降低，则货币升值；2）
若引起A国通货膨胀，CPI上升则货币贬值。

Balance-of-payments (BOP) accounting

Balance-of-payments (BOP) accounting is a method used to keep track of transactions between a country
and its international trading partners. It includes {government, consumer, and business} transactions. The
BOP accounts reflect all payments and liabilities to foreigners as well as all payments and obligations
received from foreigners.
国际收支平衡表（BOP）应该为零，这是为了与复式簿记保持一致。一方面是经常账户，另一方面是资本和金融账
户，应该相互平衡。
current account：约等于 eXports - iMports
surplus盈余表示净出口大，defict赤字表示净进口大
 financial(capital) account：looks at assets between countries.
financial account= inflows - outflows。表示对本国的net investment flows
outflows: 本国花钱购买外国资产、并持有。demestic purchases of foreign assets ==> Government-
owned assets abroad
inflows: 外国花钱购买本国资产、并持有。Foreign purchases of local assets ==> Foreign-owned
assets in the country
一个国家净出口挣的钱肯定会投资于外国资产（对应outflows）。（只是计量是用本国货币）
值> 0（inflows > outflows)，表示外国净持有本国资产、利用国外的资金来满足本国需要。Thus, the
economy is using world savings to meet its local investment and consumption demands. It is a net
debtor to the rest of the world.
值< 0（inflows < outflows)表示净持有国外资产、向国外提供资金。That indicates the economy of this
country is a net creditor, providing funds to the world.
financial account (also known as the capital account) measures the flow of funds for debt and equity
investment into and out of the country.
current account + financial account = 0

INFLUENCE OF BOP ON EXCHANGE RATES

Current Account Influences： Current account deficits lead to a depreciation of domestic currency via a
variety of mechanisms:
Flow supply/demand mechanism： 经常性账户赤字导致本国货币贬值，本国货币贬值导致进口商品变贵、出
口商品相对便宜，这会加剧经常性账户由赤字转向balance。
Portfolio balance mechanism.：一个国家Current Account surpluses 经常对应着capital account deficits
（这常常表明对国外其他几个国家的投资）。如果本国（investor country）decide to rebalance their
investment porolios, it can have a significant negave impact on the value of those investee country
currencies.
Debt sustainability mechanism：一个国家current account deficit对应着capital account surplus（by
borrowing from abroad，认为外国对本国的投资、本国欠国外外债）. 如果赤字or外债相对GDP过大，外国投
资者质疑其还债能力的可持续性，则会导致该国货币的贬值。以此来减少进口、增加出口，缩小current
account deficit，缩小capital account surplus，达到一个正常债务水平。
Capital Account Influences：

Mundell-Fleming model（蒙代尔-弗莱明模型）: 评估短期货币和财政政策对利率的影响、进而对汇率的影响。

Flexible Exchange Rate Regimes 自由汇率机制

High Capital Mobility资本高速流动。从资本账户的资本流动、货币供需分析对利率（进而汇率）的影响。

扩张的货币政策会降低利率，降低外国资本投入（reduce the inflow of capital investment in physical
and financial assets），进一步而言外国不会购买本国货币（降低本国货币的需求）、进而导致货币贬值。
扩张的财政政策会增加政府借款、增加利率，吸引国外投资，增加本国货币需求，进而导致货币升值。
Low Capital Mobility资本低速流动（有限流动），主要从经常性账户分析货币和财政政策对汇率的影响。The
effects of monetary and fiscal policy on exchange rates will operate primarily through trade flows
rather than capital flows.
扩张的货币政策或财政政策，都会刺激国内需求、增加净进口net import，则兑外币需求增加、进而导致本
国货币贬值。
The combination of restrictive monetary and fiscal policy will be bullish for a currency. This policy
mix will tend to reduce imports, leading to an improvement in the trade balance.

Fixed Exchange Rate Regimes固定汇率机制

 扩张的货币政策同样会导致本国货币贬值，政府则需要在外汇市场上购买本国货币，来维持固定汇率。这由于初
始的货币扩张政策相反。
因此一个政府若想有独立的货币政策，则必须让汇率自由浮动或者限制资本自由流动（如上面）

“不可能三角”即一个国家不可能同时实现资本流动自由(free capital flow)，货币政策的独立性(independent

monetary policy)和汇率的稳定性(有stable exchange rate)。

1999年，美国麻省理工学院教授克鲁格曼在蒙代尔-弗莱明模型的基础上，结合对亚洲金融危机的实证分析，提
出了“不可能三角”(Impossible triangle/Impossible trinity theory)
independent monetary policy: 指本国的利率调整是否要一定要和 国外世界利率一致，不然就会有套利空
间（引起资本自由流动或者汇率不稳定）
中国：要货币政策的独立性、汇率的稳定性，有资本管制（没有实现资本的自由流动）
香港：要资本自由流动和汇率的稳定性，没有独立的货币政策
欧洲国家： 有stable exchange rate and free capital flows ，没有独立的货币政策

Monetary models只考虑货币政策对汇率的影响，不考虑财政政策的影响。

Pure monetary model： the PPP（购买力平价公式） holds at any point in time and output is held constant.
货币共计增加x%、物价增加x%，货币贬值x%
Does not take into account expectations about future monetary expansion or contraction.即货币供给未
来的增加不会影响当前的汇率值。
Dornbusch overshooting model（多恩布施 超调模型）
This model assumes that prices are sticky (inflexible) in the short term and, hence, do not immediately
reflect changes in monetary policy (in other words, PPP does not hold in the short term). 假设物价短期
内不会受货币政策的影响
该模型认为货币扩张政策，在短期内不仅会由于物价增加x%带来货币贬值x%,而且还会由于资本外流capital
outflows使得domestic currency贬值大于x%。

Portfolio balance approach takes a long-term view and evaluates the effects of a sustained fiscal deficit or
surplus on currency values.（不考虑货币政策对汇率的影响）

长期的财政扩张政策，在短期会因增加政府借款、增加利率，吸引国外投资本国货币，增加本国货币需求，进而导致
货币升值。但长期政府需要归还贷款，必须紧缩财政政策（利用税收还债）或者印钱（扩张货币政策）这都会导致汇
率贬值。

政府想要干预汇率一般需要相对的足够的外汇储备。

货币危机的几个警告信号：出口减少、固定汇率、外汇储备减少、通胀增加、货币供给快速增加、银行危机

R6: Economic Growth

Growth Factors and Production Function

Preconditions For Growth

Savings and investment 与经济发展正相关

Financial markets and intermediaries 金融市场和媒介越多越高效，越利于经济增长
The political stability, rule of law, and property rights
Investment in human capital
 Tax and regulatory systems
Free trade and unrestricted capital flow

GDP, potential GDP，equity return

E(R) = dividend yield (DY) + expected capital gains yield (CGY)

expected capital gains yield = EPS growth (Δ EPS) + expected repricing (ΔP/E)
EPS growth (Δ EPS)= real EPS growth (Δ EPS_R) + inflation( π ) + change in shares outstanding(ΔS)
E(R) = DY + ΔEP SR + π + ΔS + ΔP /E​

Therefore, higher potential GDP growth implies higher real interest rates and higher real asset returns in
general.

Factors inputs and economic growth

The Cobb–Douglas production function

Y = T K α L1−α
Y the output GDP
K: capital
L: labor
T: total factor productivity全要素生产效率
α, 1 − α the share of output allocated to capital and labor
It exhibits constant returns to scale. 输入增加多少倍、产出就增加多少倍。柯布-道格拉斯生产函数呈现规模
报酬不变。
output per work(labor productivity) = Y /L = T ( K
L)
α ​

增加labor productivity，可以通过增加T和 K/L(capital per work,or capital deepening)

labor productivity growth rate
= growth due to technological change + growth due to capital deepening

Growth Accounting and Influencing Factors

the growth in potential GDP

ΔA
ΔY /Y = A ​ + α ΔK​
ΔL
K + (1 − α) L ​

A = technology, is short for T

growth rate in potential GDP = long-term growth rate of technology + α * long-term growth rate of
capital + (1- α ) * long-term growth rate of labor
* α = the elasticity of output to capital = the shares of output allocated to capital
* 1 − α = the elasticity of output to labor = the shares of output allocated to labor
Y = Y
L ​ ⋅ L , GDP = labor producitivity * labor
growth rate in potential GDP = long-term growth rate of labor + long-term growth rate of labor
productivity(output per work,)

Labor Supply Factors:

Definations:
Quantity of labor = the size of the labor force * average hours worked。
Labor force = the number of working age (ages 16–64) people available to work, both employed and
unemployed.
Demographics人口统计数据（年龄分布等）
 Labor force participation =Labor force / working age population
劳动力参与力 = 劳动力 / 适龄人口数量
Immigration
Average hours worked

Investmenting in human capital, physical capital, technologies, public infrustructure can increase economic
growth.

Growth and Convergence Theories

classical growth theory古典增长理为：

1、假设：人均收入大于可维持水平，就会生孩子、人口增长
2、因此人均GDP的增长是不可以永久的，因为人均GDP上升到生存水平以上、就会出现人口爆炸/增长。
3、Classical growth theory is not supported by empirical evidence.没有经验证据的支持

neoclassical growth theory 新古典增长理论：

1、主要关注经济长期稳态的增长率 (sustainable growth rate or equilibrium growth rate).

经济增长与当前技术水平、人均资本率capital-to-labor相关。
假设技术增长才会带来资本投资的增加。 assumes that capital investment will expand as technology
improves

2、Based on the Cobb-Douglas function discussed earlier, neoclassical growth theory states that:

θ
g∗ = 1−α ​

g* = Sustainable growth of output per capita (or output per worker)

θ : growth rate in technology
(1 – α): labor’s share of GDP
θ
G∗ = g ∗ +ΔL = 1−α ​ + ΔL
G*: Sustainable growth rate of output
ΔL: the growth of labor

3、Under neoclassical theory,

population growth is independent of economic growth.人口增长和经济增长是独立的。

Endogenous Growth Theory内生增长理论

1、技术增长来自于对physical and human capital的investment。而且Technological progress enhances

productivity of both labor and capital.
2、Under Endogenous Growth Theory:
There is no steady state growth rate, so that increased investment can permanently increase the rate of
growth.
Theorizes that returns to capital are constant, implies that an increase in savings will permanently
increase the growth rate.

Explain and evaluate convergence hypotheses.

 The absolute convergence hypothesis（绝对收敛假设）： states that less-developed countries will converge
to the growth rate (but not the level of per capita output) of more-developed countries.
means that convergence between developing and developed countries is a given, regardless of their
particular characteristics. 无论发展中或发达国家的一些特征是否相似，欠发达国家的的增长率都会收敛到更发
达国家的水平。
The neoclassical model assumes that every country has access to the same technology. This leads to
countries having the same growth rates but not the same per capita income（但per capita
output/income 不一样，和国家当前GDP基数有关）
The conditional convergence hypothesis（相同条件收敛假设）： states that convergence in living standards
will only occur for countries with the same savings rates, population growth rates, and production
functions.
具有相同储蓄率、人口增长率、产品功能的国家，其生活水平会收敛到同样的growth rate，per capita output.
此假设下，欠发达国家的增长率会更高一些，直到赶上或达到发达国家的水准。Under the conditional
convergence hypothesis, the growth rate will be higher for less developed countries until they catch up
and achieve a similar standard of living.
If these conditions hold, the neoclassical model implies convergence to the same level of per capita
output as well as the same steady-state growth rate.
The club convergence hypothesis（俱乐部呢收敛假设）：同一个俱乐部内的国家中，中等国家会较快速度发展、
直到收敛到较发达国家的growth rate和per capita output.
Under this hypothesis, countries may be part of a ‘club’ (i.e., countries with similar institutional
features such as savings rates, financial markets, property rights, health and educational services, etc.).
Under club convergence, poorer countries that are part of the club will grow rapidly to catch up with
their richer peers. Countries can ‘join’ the club by making appropriate institutional changes. Those
countries that are not part of the club may never achieve the higher standard of living.

R7: Economics of Regulation

Economic Rationale for Regulation

informational frictions, 信息摩擦、传播不畅

externalities，例如污染生成者不会为环境破坏买单。
weak competition, need Antitrust regulations(反垄断法规）)
social objectives.
上面情况出现是，需要regulation

Regulating commerce. vs Regulating financial markets.

Regulating commerce：规范商业贸易，指政府出台法律法规
Regulating financial markets：主要监管 securities markets and financial institutions.
xx
Regulation of Financial Institutions
Prudential supervision refers to the monitoring and regulation of financial institutions to reduce
system-wide risks and to protect investors.
Antitrust Regulation

Regulation can be classified as

 Statutes: reflecting laws enacted by legislative bodies 立法：反映立法机构制定的法律
Administrative regulations or administrative law: rules issued by government agencies or other regulators
行政法规或行政法：颁布的规则政府机构或其他监管机构
Judicial law: interpretations of courts 司法：法院的解释

Regulators

government agencies
independent regulators: are given recognition by government agencies and have power to make rules and
enforce them. eg. self-regulating organizations (SROs) are also independently funded and, as such, are
politically independent.

Regulatory Interdependencies

Regulatory capture：指监管机构会采纳被监管机构的一些意见，或者一些监管规定会。来做出一些保护被监管机构
的利益（如限制新进入者、保护已有行业利润）
Regulatory capture is more likely to be a concern with Self-Regulating organizations (SROs) than with
government agencies. For example, regulatory capture is often cited as a concern with the
commercializaon of financial exchanges.与政府机构相比，监管捕获更可能成为 SRO 的担忧。 例如，监管捕
获经常被认为是金融交易商业化的一个问题。
Regulatory competion: 指不同juridictions的regulator互相竞争、provide the most business-friendly
regulatory environment，目的是吸引外国公司来该地区。
Regulatory arbitrage：监管套利。指不同地区的监管规则不一样（比如污染）或者同一地区监管条纹解释和经济行
为实质之间的区别。

Tools of Regulatory Intervention

Price mechanisms
Restricting or requiring certain activities.
Provision of public goods or financing of private projects

Cost Benefit Analysis of Regulation

net regulatory burden = Regulatory burden - the private benefits of regulation

net regulatory burden <0，监管成本大于带来的收益，还不如不监管？

Comments
labor productivity = output per worker
capital-to-labor ratio = capital deepening

Covered interest parity is forced by arbitrage, which is not the case for uncovered interest rate parity. If the
forward rate is equal to the expected future spot rate, we say that the forward rate is an unbiased predictor of
the future spot rate: F = E(S1). In this special case, given that covered interest parity holds, uncovered interest
parity would also hold (and vice versa). In other words, if uncovered interest rate parity (and covered interest
parity) holds, the forward rate is unbiased predictor of future spot rate (i.e., forward rate parity holds).

broad money growth will increase.广义货币增速将加快。

institutional impediments 体制障碍

As a potential source of funds, foreign investment breaks the vicious cycle of low income, low domestic savings,
and low investment.外资作为潜在的资金来源，打破了低收入、低国内储蓄、低投资的恶性循环。

launching a rural literacy program启动农村扫盲计划

building a group of auto and textile factories in the southern states 在南部各州建设一批汽车和纺织工厂

GDP per capita，人均GDP， 有时也写为 per capita GDP

GDP per capita growth 人均GDP增长

Low per capita GDP suggests that India may lack sufficient industrial and financial infrastructure to support
some types of industries.人均GDP较低表明印度可能缺乏足够的工业和金融基础设施来支持某些类型的产业。

Blackout periods are established by companies in response to concerns about insider trading. 为了应对内幕交易
的担忧，公司设立了封锁期。

Sengupta’s endorsement of an exchange that trades “pollution rights” is consistent with the Coase
theorem. The Coase theorem states that if an externality can be traded and there are no transaction costs, then
the allocation of property rights will be efficient and the resource allocation will not depend on the initial
assignment of property rights.