Econometrics - Forecasting Recession Probabilities via Large Factor Model Regression

Anav Sahni

Department of Economics – Faculty of Business, Coventry University

May 2020

Abstract

This thesis estimates the probability of a recession within the near term through binary Probit Regression
Modeling. It will predict the likelihood of a future recessionary period using quarterly data from Q1 1955 to
Q1 2020 of 212 economic and business indicators. The binary regression model extends Stock and Watson's
(1992) large factor model predictor of 160 financial indicators by expanding the variable list and the time
series included. Methodological and statistical considerations are critically evaluated while constructing the
most comprehensive recession forecasting model dataset thus far. Acknowledging the absence of
optimization, it is the significance of the model's forecasting predictions that invites subsequent testing and
development. The large factor model indicates the real headwinds are yet to be experienced and at the end of
the current market rebound, a downward trajectory is shown.

JEL classification: Y40

Key words: Regression, Quantitative Methods, Econometric Forecasting.

 Certificate of Ethical Approval
Applicant:

Anav Sahni

Project Title:

Forecasting recession probabilities with a large set of macroeconomic Indicators

This is to certify that the above named applicant has completed the Coventry
University Ethical Approval process and their project has been confirmed and
approved as Low Risk

Date of approval:

04 March 2020

Project Reference Number:

P103352

2
 Table of Contents:

• Abstract………………………………………………………………………………………………………….….……..….…….………….1

• Ethics Approval………………………………………………………………………………………….…….….……..….…….…………2

• Table of Contents…………………………………………………………………………………….…….….……..….…….……….….3

• Acknowledgements………………………………………………………………………………..........….……..….……...........…4

• Introduction ……………………………………………………………………………………...……….….….……..….…….………….5

• Literature Review ………………………………………………………………………………………….…..……..….…….………….6

• Methodology………………………………………………………………………………………………….…..……..….…….………...8

• Limitations………………………………………………………………………………………………………..….……..….…….……...10

• Data Collection ………………………………………………….………………………………………………………………..….…….11

• Results and Analyses………………………………………………….…………………………….….……..….….….…….….…….12

• Conclusion…………………………………………………………………………………………….….……..….…….….………………20

• Appendix 1 – Factors List………………………………………………………………...….……..….…….……………….…….…21

• Appendix 2 – Regression Results………......……………………………………………………...….………… ……….………24

• Appendix 3 – Recession Regression Model Code ……….............……………………...….………………….………28

• References………………………………………………………………………………………...….……..….…….………………..……35

3
 Acknowledgements

I would like to thank [Link] Nissah, [Link] Rohith, and [Link] Carruthers from the School of
Economics at Coventry University for their support and guidance throughout this study. I would like to thank
the Coventry University Faculty of Business for the extensive resources and facilities contributed to this study.
Further, I would like to thank students Rhea Prashar, Arjun Jaitha, and Olayemi Ogundele for their input in
the study. Lastly, kind regards to my sister, parents, friends, and family for their unwavering support and
encouragement throughout. Any errors or omissions remain the author’s responsibility.

4
 [Link]

The historical nature of recessions has always been characterized by the underlying element of surprise. As a
result of the Great Depression, business cycle research largely contributes to the evolution of modern
econometrics and macroeconomics. As economists, we have witnessed tremendous maturity in forecasting
models yet incapable of accurately predicting and preventing the last global recession. The historical
challenges and inadequacies in forecasting form the primary incentive to embark on the present study.

Previously, econometrists have created an entire scope of models with a variety of factors, time-periods, and
optimization & regression methods. These include but are not limited to, highly optimized small factor models
with widespread time-periods, generalized large factor models of specific periods, complex regression
methods, and countless others. Upon examination of seminal work through the ages, a peculiarity in research
has manifested. Notable work presents a failure to firstly include data from the last nine recessionary periods
with no convergence on indicators and weighted variables. These hurdles and variations provide scope for
alternative theories and forecasting modifications that can provide a degree of reliability and accuracy. The
07' recession exemplifies the domino effect of a collapse in the overleveraged mortgage sector and the ensuing
impact in the US and world economy. It is the fall in the common denominator, the universal dollar, that ties
the global economies to aggregate national shocks. Increasing focus on the bond yield spreads and curve
inversions pre-recession have spawned studies targeting bond yields as a leading indicator. The inversion of
the US bond yield spreads first occurred in December 2018, marking the first inversion since the great
recession, and sounded the alarms for what could ensue. It is not the inversion rather the reversion of the
spread that is recorded, in anticipation of a recession. It is the disparity in months preceding to a downfall that
is most telling of the current landscape as we are 2 quarters away from meeting the average months to recession
after inversion. With a threat of recession looming, the US interest rate cuts, rising global trade tensions, and
huge finance & banking pressure, the impact of predictability and in turn forecasting, will be of great
importance in strategies such as policy setting, financial management, business realignment, while providing
a suitable response time for public and private sectors.

This study intends to estimate the probability of a recession in the near to short term and whether the creation
of the potentially largest model can boost predictability and forecasting power. The study estimates this
likelihood, by encompassing the longest time-period in a recession forecasting model to provide a degree of
statistical accuracy in the results. The primary driver undertaking this study is not only the indisputable
relevance in modern economics but also the aforementioned state of the influential US economy and the
heightened global pressure and uncertainty. The present global climate is one that has endured the highest
levels of uncertainty, trade tensions, leveraged debt, and OPEC incongruities that could trigger a recession.
Theoretically, an economic downfall will show in declining market data with some indicators lagged behind
others. Since it is near impossible to predict a future or current trigger, a comprehensive model with over 200
indicators has been built to include the following groups (8). Money, banking & finance, population,
employment & labor markets, national accounts (USA), production & business activity, prices, international
data, regional data (USA), and academic data. The large factor modeling approach is not novel, however,
observing the success of seminal authors Stock and Watson's (2006) paper using 160 variables has led to the
creation of a model with more depth and breadth in the data set.

5
 [Link] Review

The study of modern econometric analysis of business cycles dating back to the early '30s became largely
popular in pursuit of a greater understanding of business cycle fluctuations preceding the great depression and
world war II. Here is a concise review focusing only on the history of econometrics in business cycle modeling
and the turning points in methodology through influential work.

Deemed the father of econometrics, Jan Tinbergen was one of the first economists to create macro-
econometric models. Tinbergen's (1939) book "Business cycles in the United States" garnered high praise for
the first of its kind, multi-equation dynamic macro-econometric model of business cycles that received a Nobel
Prize. His instrumental role in the history of econometrics was influential to the foremost economists Ragnar
Fisch, J.M Keynes, and Milton Friedman.

Tinbergen's study and debates over methodology provided the necessary stimuli for the Haavelmo and the
Cowles commission research group in the 1930s to formalize and advance the study of econometric business
cycle modeling. Haavelmo's probability approach in the early '40s was pioneering to the foundations of
probability theory in econometrics. While their work was influential, it failed to resolve the methodological
issues that earlier work raised and fell under heavy criticism. An extensive methodological review of the
business cycle research up to the '50s can be read in Koopman's (1949) paper 'The econometric approach to
business cycle fluctuations', which concluded that business cycles are embedded in the dynamics of certain
macroeconomic variables such as GNP & GDP, that are driven by a few aggregate variables plus random
shocks. Following this methodology, econometrists set out to obtain the best statistical estimates for the
coefficients of structural models, and once found, only then an explanation of the model was presented.

Recognizing this fundamental flaw, the NBER (National Bureau of Economic Research) sought to address
the issues in their business cycle studies. Under the leadership of Wesley Mitchell, the NBER put forward a
mature procedure in establishing empirical evidence of dating business cycles by their troughs to peaks in
individual time series of France, Germany, the US, and the UK in efforts to formalize business cycle studies.
Mainstream macroeconomic modeling up until the '60s diverted towards static economic theories looking at
simultaneous equation models (SEM) framework rather than business-cycle studies.

Only by the 1973 oil crisis led recession was the focus on business cycle research revitalized. This rejuvenation
led to the focus on dynamic features and time-series properties of macro-econometric data and models.
Consequently, business cycle research evolved to compute the short-run dynamics, relying profoundly on
computer-simulated theoretical models. The first strand of the dynamic modeling approach was undertaken
by Sargent and Sims (1977) in their study, "Business cycle modeling without pretending to have too much a
priori economic theory", that adapted the NBER method of business cycle analysis by using vector
autoregression technique (VAR) on 14 time-series variables using quarterly data from 1949 to 1971, in efforts
to forecast the '74 recession using VAR. This led to the rise of the VAR approach that amassed wide criticism
and skepticism of its theory. A concise review of the history of the vector autoregressive theory has been
written in Qin's (2008b) paper "Rise of VAR modeling approach".

That brings us to the end of the '70s, after which, a new strand of theory, known as the Real Business Cycle
(RBC) approach, was instated. The RBC approach conscripted by Kyland and Prescott (1982) in their study
"Time to build and aggregate fluctuations" examined real factors, rather than nominal factors as they assumed
business cycle features were incorporated in the autocorrelation of real GDP and its covariants, that led to the
simulation of cyclical features as the fundamental objective. Kyland and Prescott followed the 'computable
general equilibrium' model (CGE) and developed it into the 'dynamic stochastic general equilibrium' known
as the DSGE model that still serves as a reference for evaluating and adjusting model simulation results. The
history of CGE models can be found in Mitra-Kahn's (2008) paper "Debunking the myths of CGE models'.
DSGE models went further as econometrists set targets for their RBC models by mimicking simulations of
time-series properties of aggregate variables.

6
The 80s saw rapid formalization of NBER business cycle measures by time-series econometrics as economists
focused on trend following for the non-stationary variables. Simultaneously, a small group at the NBER
focused on automating the process of selecting a binary series of peaks and troughs for turning point
identification. This led to the proposal for discrete-state Markow process in single time-series models for
forecasting turning points in macro variables (e.g unemployment) by Neftci (1982) in the "Optimal prediction
of cyclical downtowns" paper, which was taken further by Hamilton (1989) in his innovative paper, "A new
approach to the economic analysis of nonstationary time series and the business cycles".

At the same time, Stock and Watson (1989) resumed the research devised by Sargent and Sims (1977) on
"modeling without priori knowledge" as aforementioned, by exercising an experimental approach using factor
analysis of multivariate forecasting. Stock and Watson employed a dynamic factor model (DFM) to create a
single coincident index published in 1989, from the NBER variable list looking at the first lad of trended data
series similar to Hamilton's 1989 work.

Stock and Watson later found that their choice of specific leading indicators was inadequate, resulting in
mispredictions, as shown in their 1993 "econometric issues of predicting recessions" paper in the book
'Business Cycles, Indicators, and Forecasting' by UChicago. The results highlighted the significance of
recognizing "timely particular shocks" that would cause non-periodic business cycles by tracking the source
to a combination of small and large shocks, rather than small shocks depicted by the Slutsky-Fisch impulse
propagation scheme formerly mentioned. These findings played a pivotal role in motivating Stock and
Watson's research into DFM while other economists developed different statistical models. The era saw a
rapid growth of exploratory models such as error correction models, theoretical DSGE models, mixed models,
and experimenting with probit models to focus on forecasting the probability of turning points that led to
increased attention.

S&W(2006) went on to use their leading indicator (1989) of 160 variables for recession probability modeling
in their pioneering study, 'Forecasting with many Predictors', discovering that it is a handful of structural
shocks that cause the comovements among macro variables at all lags that might be as low as 2. This is
supported in canonical work by Sargent and Sims (1977), Sargent (1989), Quah and Sargent(1992), and more
recently Giannone, Reichlin, and Sala (2004). S&W also established that a large "n" in forecasting, referring
to their comprehensive set of variables, "is a blessing" in predictability. This is continually followed in recent
DFM studies by Ding and Hwang (2001), Formi, Lippi, Hallin, Reichlin(2001), and Bai & Ng
(2002,2004,2006); finding success in large forecasting models as well. That concludes the notable literature
on econometric studies of business cycles and recession probability studies this concise review has covered.
An extensive report can be read in Qin's (2010) paper, "Econometric studies of business cycles in the history
of econometrics".

The modern econometric community is currently divided by mathematical fundamentalists over rigor and
consistency. It is often the case that mathematical accuracy is the dominant goal currently, and the relevance
of research to recession probabilities is merely diminished to observational illustrations. The last economic
downturn, the great recession, went largely undetected by forecasters and econometrists. That in itself reminds
us of the complexity in building an accurate model to perfectly track future economic activity. The history of
econometrics tells us that new discoveries would not occur until all possible paths have been traversed and
that shouldn't stop advancing those boundaries.

7
 [Link]

The forecasting model employed in this study pursues the binary limited dependant variable (LDV) model. A
limited dependant variable model estimates the probability of the dependant variable Y, using qualitative
independent variables Xi. The estimation technique operated in the model is a binary approach known as
Probit Regression, primarily used to overcome the limitations of a linear probability model.

The model will follow binary specifications as y = 0 indicates a recession and Y = 1 represents the opposite,
non-recession. Hence, the estimated value of dependant variable Y is the probability of state y=0, a recession,
at given levels of X. A probit regression model encompasses the cumulative normal distribution within the
range of 0 and 1 and seeing as most economic indicators are normally distributed, probit regression is generally
preferred. Conscious of the similarities in results and marginal effects between a binary logit and binary probit
models, significant evidence highlight the difficulties in interpreting large logit models.

In a probit model, the cumulative distribution function (CDF), Φ(Phi) is defined within the range of 0 and 1.
That is for a given Z score, Φ (Z) ∝ 𝟎, 𝟏

where
𝑝𝑖 = 𝑝(𝑦𝑖 = 1) = Φ (𝑍𝑖 ), and

𝑍𝑖 = Φ−1 (𝑦𝑖 = 1) = 𝛽0 + 𝛽1 𝑋𝑖 + 𝜀𝑖

Φ (𝑍𝑖 ), is the normal cumulative density function where the area under the normal distribution curve for a
random variable z is the probit link function for which the formula for Φ(Z) is
2
1 𝛽0 + 𝛽1 𝑋𝑖 + 𝜀𝑖 −𝑠
𝑝𝑖 = 𝑝(𝑦𝑖 = 1) = ∫ 𝑒 2 𝑑𝑠
√2𝜋 −∞

This is the fundamental teaching behind probit regression modeling. Implementing a large number of
indicators mimicking Stock and Watson's (2006) study will result in a simple expansion of the equation to
include more Xi variables.

Extending the aforementioned framework by Stock and Watson (2006) by regressing a larger dataset and time
series model than previously instated will provide new estimates whilst sacrificing math rigor and consistency
in this experimental econometric model. The study seeks to replicate a limited dependant variable binary
model by using panel data of 212 economic and business indicators across 8 groups. These being: Money,
Banking & Finance (44), Population, Employment & Labor Markets (62) , National Accounts(53), Production
& Business Activity (31), Prices(29), International Data(1), US Regional Data (4), and Academic Data(1). In
practice, large factor models have found forecasting success in the past, evident in works by Christiansen et
al (2014), Chen, Iqbal & Lai (2011), and Henri Nyberg (2010), who also preceded this approach. Hench, there
is strong founded intuition that the model may have a degree of predictability, seeing as it incorporates and
fits a very large amount of data collected from the Federal Reserve's Economic Database. The corresponding
8 groups of variables comprised of 212 indicators utilized in this study, is composed of quarterly data from
Q1 1955 up until Q1 202. The complete list of indicators can be found in the data collection section in the
appendix. The model attempts to forecast the probability of a recessionary period in the United States as it is
the largest economy, a data-rich econometric environment, and a global common currency.

Before executing the regression, a correlation coefficient analysis on each group using the Pearson correlation
coefficient must be computed to determine how the independent variables are related to our binary dependant
variable and each other. Shadowing the work of seminal papers using lagged variables of independent
variables, this study will also illustrate the recession probabilities up to 3 quarters ahead. The Pearson
correlation coefficient will highlight the few inadequate indicators to be removed to avoid serial

8
autocorrelation. Thereafter, systematically omitting variables due to missing variables or statistical errors in
their data or relevance.
The research and regression will be performed on each individual group and then the whole model to forecast
probabilities in future time periods. The research will test the forecasting abilities using up to 3 lags ahead to
compare and any highlight significant group-wise turning point in the large model between lags.

After retrieving the data and graphs from STATA, the study can be taken further by superimposing
significantly correlated indicators onto the binary variable, US economy.

This may be the best approach for this large factor model as binary probit modeling firstly follows the 0 to 1
range like the probability scale. Furthermore, previous studies also incorporated large factor models using
probit modeling due to its cumulative normal distribution shape. Lastly, while the probit modeling approach
is relatively easier than the dynamic model approach, it is a combination of experiential reasoning and growing
econometric skills that led to its use.

The research undertook shadows the work of Stock and Watson by expanding its variable list and using the
probit regression technique for the large panel data, similar to recent works by Nyberg (2010), Chen, Iqbal,
and Lai (2011), as well as Thomas Hsu (2016). The model will add to the seminal work by not only using
more factors but also using a different regression model than originally instated. The reasoning behind the
model has formerly been mentioned, whereas the justification of increased variables stems from the risk of
missing a 'handful of timely particular shocks' that Stock and Watson eluded to as the reason for the great
recession.

9
 [Link]

As aforementioned, the use of probit models is well suited in handling random coefficients of variables that
are normally distributed. In this regard, the only limitation exhibited by probit models is the requirement of
normal distribution for all unobserved components of the equation. In its absence, normal distributions from
the model can become redundant and even hinder forecasts. Undertaking this probit forecasting model led to
a batch of restrictions uncovered in the data, rather than the model.

Due to the sheer volume of data points, the forecasting model overcomes a series of minor issues under
development. The initial creation of the panel data comprising of 212 indicators was problematic due to
disparities in each data series that had to be normalized. These incongruities arose from variances in reporting
units, frequencies of releases, and missing complete data from 1955. While these data limitations can be
proxied using backcasting and optimization methods, leading backtesting model econometrics are as complex
as leading forecasting models deeming them admissable for this study, however, providing scope to take this
model forward. The reliance on manipulated data was not an applicable route, seeing as raw data will continue
to be used as this model progresses making sure to not influence the results of the model in any way. Formerly
explained in the methodology, a primary driver towards the use of probit regression was the ability to
accommodate missing data and random substitution.

Latterly, the utilization of a large data set for modeling connotes the possibility of an overfed and subsequently,
an overfitted forecast. Moreover, large factor approaches can be hindered by error correlations, serial
autocorrelation, and/or large covariances among indicators, that may cause statistical or mathematical
redundancies of the model as a whole. Most limitations were overcome during the development of this model
while remarking the difficulty in finding a good fit. Using a probit model in comparison to VAR, DSGE, and
more dynamic modeling approaches may not be mathematically superior, however as a limited dependant
variable model, using these dynamic approaches over this proposed time-series may not prove favorable in
long-run forecasting, nor applicable in the case for large factor analysis.

10
 [Link] Collection

The dataset used in this model has been compiled using the Federal Reserve Economic Data maintained by
the Federal Reserve Bank of St. Louis database. There is a total of 212 factors in the dataset stating from Q1
1955, to include the last 9 recessionary periods. The data frequency is reported in quarters and can be classified
by the following groups:

[Link], Banking & Finance 2. Population, Employment & Labor Markets 3. National Accounts
[Link] & Business Activity [Link] [Link] Data [Link] Regional Data 8. Academic Data

A complete list of factors can be viewed in chronological group order in Appendix 1.

11
 6. Data & Results Analyses

The regression results exhibited from the probit model are presented in appendix 2. Noting that the total list
of variables had a 10% omission of variables due to high serial autocorrelation and overly large covariance
with the dependant variable. Taking account of the piecewise regression data, it is important to recognize the
regression results for larger groups aren't as clear as the regression results for small groups. This boils down
to the sample sizes and limitations in optimization methods, or lack thereof, in comparison to leading models.
Naturally, findings highlight how individual group regression results are more promising than the consolidated
group regression in the predicted probabilities as well.

The probability of a recession changes across time with respect to each group (1 to 8) emphasizing how the
probability of future recession at certain time periods is seemingly higher(or lower). This denotes how a
contrary change in the leading indicators will have an effect on the recession probability. Carrying out this
analysis with lagged variables in groups 1-8 to recognize the evolution of the prediction over 3 lags in the
future. We regress up to 3 lags in order to forecast the recession probability over the next 3 periods.

Looking further, the predicted probability graphs show how the probability of a recession changes over time
with respect to each group of indicators and as then for the large model as a whole. From the data below, it is
interesting to see how the smaller models, exercised in groups 6,7,8 provide better results than the bigger
group models and the large model as a whole. Recounting group 8 forecasts a high probability of a recession
while others do not exhibit a detrend as such. We can see how it unfolds in the lagged predictions ahead.

12
 Large Model

Lagged Forecasts
Next, we look at the first lagged forecast prediction graphs of each group and the whole sample below
representing the forecast 1 quarter ahead. Reporting the regression results for each lag is considered
unnecessary due to the illustrative purpose of probit’s turning point predictions. Seeing as the data is in the
quarterly format, each lag represents 1 quarter and thus forecasting the 3,6,9-month forecast horizon below.
Once again, witnessing no signs of an oncoming recession in any of these forecasts. Intriguing results of the
first lag highlight group 8 no longer exhibits an inflection on the graph.

13
 Large Model

The next forecast presented below exemplifies the second lag of the forecast graphs. While most groups and
the large model are relatively stationary, note how groups 6 and 7 indicate an economic downturn foregone
while group 8 illustrates a slower de-trended data to come, hinting the likelihood of a recession.

14
 Large Model

The last and final recession probability forecast graphs illustrated is the 3rd lag shown below. Once again,
witnessing a similar portrayal of the recession forecasts in the 2nd lag. It can be said that the smaller groups
hold more predictive power, seeing how the probability changes with each variable over time. With the larger
groups, however, this can be very technical and mathematically heavy, leading to improper results or model
grouping to find the best fit when there is maybe none. This leads to larger models being able to predict
recessions more “accurately” when we know this is may not always be the case.

15
 Large Model

While an enhanced portrait is drawn via individual groups' forecasts, higher-level analysis on the consolidated
data can be pictured by superimposing significantly correlated indicators onto the binary variable, US
recession over a time series plot. Providing greater insights into what we may have already, or will incur, in
the next three quarters. The blue line represents the presence of a recession (0) and 1 representing non-
recessionary periods.

From the graph above, a clear picture is painted when looking a few quarters ahead. At the time period of this
study (Q1 2020), we can see a forecasted downturn in the future periods from the line ‘A014RE1Q156NBEA’
representing the large model’s adjusted forecast.
16
 Forecast Comparison Vs US Markets

Witnessed

2020-2021 Forecast

Focusing on the adjusted, model prediction graph of future economic recessions, it is fair to assume that the
sharp decline towards 0 (recession) can be seen, accompanied by rapid recovery, and subsequent downfall
once more. At the present time concluding the study, various arguments have been presented whether or not
the economy has been in and out of a recession, whether it is currently in one, or whether one even existed.
Taking a look at the following real stock and bond market index data of the S&P 500, Dow Jones Industrial
Average, NASDAQ composite, and Bank of America’s US High Yield Index. The emergence of a pattern is
recognized, similar to that in the prediction graph. This pattern is a downward trend at the end of the first
rebound, shown at the end of the graph. Implying that the current downfall, as news represents, may just be
investors panic selling amid coronavirus fears. The implication is that the worst is yet to come and that the
recession is yet to properly ensue if these indices follow the near-term downward pattern our model suggests.

(S&P 500, 2020)

17
 (DJIA 500, 2020)

(NASDAQ Composite Index, 2020)

(ICE BofA US High Yield Index Option-Adjusted Spread, 2020)

Contrastingly, we can analyze the results and correlated market data from the opposing view to say that we
either misinterpreted this downturn as an event-based, coronavirus triggered market tumble, or a short market
decline leading to a recovery in continuation of the general bull market.

Evaluating the results on the forecasted recession probabilities has proven to be useful in understanding the
transitory phase the US business cycle is currently undergoing. Taking account of the superimposed forecast
comparison against current US markets shows the forecasting success of the model in the past and future,
albeit limited. It is the forecasted trend that highlights the increased likelihood of a recession over the coming
months, which if correct may see the subsequent fall in economic activity over the next 2 quarters or more.
Most recessionary periods were traced to a trigger factor. Judging by the current time-period of this study and
the market trends, this trigger factor may be the global coronavirus pandemic, halting businesses, and
economic activities that will likely spur into the next global recession in lockdown.

While the abundance of data in this study may prove difficult to accurately compare with recent strands of
literature, modern econometric forecasters have published an influx of business cycle studies eluding to the
next recessionary period as early as last summer. Group forecasts at different lags of the model above also
indicate the possibility of an incoming recession while also registering foregone possibilities of an elapsed
downfall. Evaluating the extent to which we can forecast recession probabilities via the largest factor model
has been answered fully, as findings show success in gathering insights towards predicting the previous
recessions in group level and consolidated data. While the model is far from being a statistically robust and
groundbreaking model, there is still scope to take this model further by cleaning and optimizing the data and
then attempting to code a complex multi-variate factor model as this study was limited by its capabilities of a
probit model.

18
Nonetheless, the model and results can be taken further in a policy setting environment like the central bank
to provide statistical reasoning over expansionary and contractionary policies. Moreover, these results can be
used in financial institutions, credit, risk & debt markets, business management strategies, and various
practical applications. This model can also be further developed into an algorithmic equation index such as
the 'Coventry University Recession Probability Index' comparable to the LSE index that exhibits general risk
and recession metrics using real-time data. Last and most promisingly, this model can be used for fixed income
trading by prompting buy and sell signals based on the market trend. Evaluating the extent to which forecasting
recession probabilities via the largest factor model, has been answered fully, as findings confer success in
gathering insights towards predicting the previous recessions in group level and consolidated level as well as
indicate the future trend. This model can be coded to provide trade signals for quantitative investing in indices
and economy linked instruments.

19
 [Link]

Conclusively, this paper has discovered that there is a looming presence of a recession that exists in the short
term period at varying lags. Graphical trends imply a further, negative spiral than currently undergone(can go
more into detail by further summarising graphical data of groups vs large model in its various subsets). The
model, however, has succumbed to its minor limitations by exhibiting largely vague regression results due to
the large model size in absence of the calibration methods. The regression results cannot solely justify the
probability of a recession, nonetheless, it gives forecasters an overarching idea of the economies' downside
trend. It is understood that in the creation of the model, a downturn is not reliant on a sole trigger factor that
can understate or overstate the potential and magnitude of a shock. This questions the impact of a large
magnitude shock on the recession probability seeing as it may or may not be limited to 1 variable and
questioning if it will get picked up by the model. The forecasts merely indicate the likelihood. A forecast can
never definitively classify a recession as recession time-period and dynamics aren't standardized. For this
reason, following the dynamic peak to trough study using the same factor data might bring forth the expected
results proposed by Neftci's (1982) paper "optimal prediction of cyclical downturns. Solutions for the
limitations of our model has been covered earlier, however, the data and model suggest increasing the
complexity and using vector autoregression techniques for an elaborative approach meeting the mathematical
rigor and consistency needed for a leading forecasting model.

Advancements in this field allow novel econometric studies to persist in a rich data environment, similar to
the case in this paper, where the reliability and source of the data are more important than the completeness
and frequency in releases. Moreover, the dataset collated relied on the St. Louis Federal Reserve and therefore
the presence of improper sampling criticism and data inaccuracies do not persevere.

This paper has been rational and overly critical of the undertaken study given the extreme levels of scrutiny
econometrists face when publishing a breakthrough regression model. While it made small strides and
overcame a series of limitations, building an econometric forecasting model is challenging due to its
mathematical resilience needed when dissecting forecasting models. Model criticism and mathematical
astuteness have admittedly played a huge role in overcoming the boundless econometric learning curve and
calls for this model to be taken forward. By advancing this model and developing it into an algorithmic
equation index, it can be further assessed as a model while using real-time data. Evaluating the extent to which
forecasting recession probabilities via the largest factor model, has been answered fully, as findings confer
success in gathering insights towards predicting the previous recessions in group level and consolidated level
as well as indicate the future trend.

Looking at the results of the S&P 500, Dow Jones Industrial Average, NASDAQ composite, and Bank of
America’s US High Yield Bond Index shows that most models are predicting similar results at the moment
with index performance mimicking the general shape of the current business cycle fluctuation. Due to global
pandemic fears, a long term lockdown would almost guarantee a recession as economic activity is halted.
Nonetheless, the national debt saw a huge rise through stimulus packages and the Federal Reserve acting as a
buffer. The unhinged use of quantitative easing will also show temporary growth and recovery in the medium
term, with the cost of inflation and depreciation in the long run. This could only exacerbate the current
situation or the next recession as the US economy will have exhausted its fiscal and monetary options by the
end of this first recovery and result in this market crash, much worse than the great recession. In the wake of
high market volatility, it is important to know that these predictions may not be an accurate depiction of the
rest of 2020, rather a depiction of the current business cycle fluctuation.

20
 Appendix 1 – Complete Variable Lists

The corresponding data is the FRED Mnemonic code and the reporting value respectively. Variable data has
been collated individually. Index scale variables and chained dollar variables, i.e. dollar at a certain year, had
to be updated to maintain 2012 as the base year for stability and statistical similarity. *For transparency, less
than 40% of variables do not begin in 1955. With less than 10% of variables being omitted due to data
limitations.
FRED CODE: FRED Instrument: FRED VALUE: Group:
FEDFUNDS Effective Federal Funds Rate Percent 1
TB3MS 3-Month Treasury Bill: Secondary Market Rate Percent 1

TB6MS 6-Month Treasury Bill: Secondary Market Rate Percent 1

GS1 1-Year Treasury Constant Maturity Rate Percent 1
GS10 10-Year Treasury Constant Maturity Rate Percent 1
MORTGAGE30US 30-Year Fixed Rate Mortgage Average in the United States Percent 1
AAA Moody's Seasoned Aaa Corporate Bond Yield Percent 1
BAA Moody's Seasoned Baa Corporate Bond Yield© Percent 1
BAA10YM Moody's Seasoned Baa Corporate Bond Yield Relative to Yield on 10-Year Treasury Constant Maturity© Percent 1
TB6SMFFM 6-Month Treasury Bill Minus Federal Funds Rate Percent 1
T1YFF 1-Year Treasury Constant Maturity Minus Federal Funds Rate Percent 1
T10YFFM 10-Year Treasury Constant Maturity Minus Federal Funds Rate Percent 1
CPFF 3-Month Commercial Paper Minus Federal Funds Rate Percent 1

AMBSLREAL Real St. Louis Adjusted Monetary Base Index 2012=100 1

IMFSL Institutional Money Funds Billions of Dollars 1
M1REAL Real M1 Money Stock Index 2012=100 1
M2REAL Real M2 Money Stock Index 2012=100 1
MZMREAL Real MZM Money Stock Index 2012=100 1
BUSLOANS Commercial and Industrial Loans, All Commercial Banks Billions of U.S. Dollars 1
CONSUMER Consumer Loans at All Commercial Banks Billions of U.S. Dollars 1
NONREVSL Total Nonrevolving Credit Owned and Securitized, Outstanding Billions of Dollars 1
REALLN Real Estate Loans, All Commercial Banks Billions of U.S. Dollars 1
REVOLSL Total Revolving Credit Owned and Securitized, Outstanding Billions of Dollars 1
TOTALSL Total Consumer Credit Owned and Securitized, Outstanding Billions of Dollars 1
DRIWCIL Net Percentage of Domestic Banks Reporting Increased Willingness to Make Consumer Installment Loans Percent 1
VXOCLS CBOE S&P 100 Volatility Index: VXO Index 1
TWEXAFEGSMTH Trade Weighted U.S. Dollar Index: Advanced Foreign Economies, Goods and Services Index 2012=100 1
EXUSEU U.S. / Euro Foreign Exchange Rate U.S. Dollars to One Euro 1
EXSZUS Switzerland / U.S. Foreign Exchange Rate Swiss Francs to One U.S. Dollar 1
EXJPUS Japan / U.S. Foreign Exchange Rate Japanese Yen to One U.S. Dollar 1
EXUSUK U.S. / U.K. Foreign Exchange Rate U.S. Dollars to One British Pound 1
EXCAUS Canada / U.S. Foreign Exchange Rate Canadian Dollars to One U.S. Dollar 1
UMCSENT University of Michigan: Consumer Sentiment Index 2012:Q1=100 1
TOTRESNS Total Reserves of Depository Institutions Billions of Dollars 1
NONBORRES Reserves of Depository Institutions, Nonborrowed Millions of Dollars 1
GS5 5-Year Treasury Constant Maturity Rate Percent 1
TB3SMFFM 3-Month Treasury Bill Minus Federal Funds Rate Percent 1
T5YFFM 5-Year Treasury Constant Maturity Minus Federal Funds Rate Percent 1
AAAFFM Moody's Seasoned Aaa Corporate Bond Minus Federal Funds Rate Percent 1
INVEST Securities in Bank Credit at All Commercial Banks Billions of U.S. Dollars 1
DCPF3M 3-Month AA Financial Commercial Paper Rate Percent 1
NASDAQCOM
DJIA NASDAQ
Dow Jones Composite Index©
Industrial Average Index
Index, Not Feb 5, 2012=100
Seasonally Adjusted 11
FREDSP500
CODE: S&P 500 FRED Instrument: Index,FRED
Not Seasonally
VALUE: Adjusted 1
Group:
OUTNFB
DJIA Nonfarm
Dow JonesBusiness Sector:
Industrial Real Output
Average Index,Index 2012=100 Adjusted
Not Seasonally 21
OUTBS Business Sector: Real Output Index 2012=100 2
OUTMS Manufacturing Sector: Real Output Index 2012=100 2
PAYEMS All Employees: Total Nonfarm Payrolls Thousands of Persons 2
USPRIV All Employees: Total Private Industries Thousands of Persons 2
MANEMP All Employees: Manufacturing Thousands of Persons 2
SRVPRD All Employees: Service-Providing Industries Thousands of Persons 2
USGOOD All Employees: Goods-Producing Industries Thousands of Persons 2
DMANEMP All Employees: Durable Goods Thousands of Persons 2
NDMANEMP All Employees: Nondurable goods Thousands of Persons 2
USCONS All Employees: Construction Thousands of Persons 2
USEHS All Employees: Education and Health Services Thousands of Persons 2
USFIRE All Employees: Financial Activities Thousands of Persons 2
USINFO All Employees: Information Services Thousands of Persons 2
USPBS All Employees: Professional and Business Services Thousands of Persons 2
USLAH All Employees: Leisure and Hospitality Thousands of Persons 2
USSERV All Employees: Other Services Thousands of Persons 2
USMINE All Employees: Mining and logging Thousands of Persons 2
USTPU All Employees: Trade, Transportation and Utilities Thousands of Persons 2
USGOVT All Employees: Government Thousands of Persons 2
USTRADE All Employees: Retail Trade Thousands of Persons 2
USWTRADE All Employees: Wholesale Trade Thousands of Persons 2
CES9091000001 All Employees: Government: Federal Thousands of Persons 2
CES9092000001 All Employees: Government: State Government Thousands of Persons 2
CES9093000001 All Employees: Government: Local Government Thousands of Persons 2
CE16OV Civilian Employment Level Thousands of Persons 2
CIVPART Civilian Labor Force Participation Rate Percent 2
UNRATE Civilian Unemployment Rate Percent 2
LNS14000012 Unemployment Rate: 16 to 19 years Percent 2
LNS14000025 Unemployment Rate: 20 years and over, Men Percent 2
LNS14000026 Unemployment Rate: 20 years and over, Women Percent 2
UEMPLT5 Number of Civilians Unemployed for Less Than 5 Weeks Thousands of Persons 2
UEMP5TO14 Number of Civilians Unemployed for 5 to 14 Weeks Thousands of Persons 2
UEMP15T26 Number of Civilians Unemployed for 15 to 26 Weeks Thousands of Persons 2
UEMP27OV Number of Civilians Unemployed for 27 Weeks and Over Thousands of Persons 2
LNS13023621 Unemployment Level: Job Losers Thousands of Persons 2
LNS13023557 Unemployment Level: Reentrants to Labor Force Thousands of Persons 2
LNS13023705 Unemployment Level: Job Leavers Thousands of Persons 2
LNS13023569 Unemployment Level: New Entrants Thousands of Persons 2
LNS12032194 Employment Level: Part-Time for Economic Reasons, All Industries Thousands of Persons 2
HOABS Business Sector: Hours of All Persons Index 2012=100 2
HOAMS Manufacturing Sector: Hours of All Persons Index 2012=100 2
HOANBS Nonfarm Business Sector: Hours of All Persons Index 2012=100 2
AWHMAN Average Weekly Hours of Production and Nonsupervisory Employees: Manufacturing Hours 2
AWHNONAG Average Weekly Hours of Production and Nonsupervisory Employees: Total private Hours 2
AWOTMAN Average Weekly Overtime Hours of Production and Nonsupervisory Employees: Manufacturing Hours 2
IPDBS Business Sector: Implicit Price Deflator Index 2012=100 2
CES2000000008 Average Hourly Earnings of Production and Nonsupervisory Employees: Construction Dollars per Hour 2
CES3000000008 Average Hourly Earnings of Production and Nonsupervisory Employees: Manufacturing Dollars per Hour 2
COMPRMS Manufacturing Sector: Real Compensation Per Hour Index 2012=100 2
COMPRNFB Nonfarm Business Sector: Real Compensation Per Hour Index 2012=100 2
RCPHBS Business Sector: Real Compensation Per Hour Index 2012=100 2
OPHMFG Manufacturing Sector: Real Output Per Hour of All Persons Index 2012=100 2
OPHNFB Nonfarm Business Sector: Real Output Per Hour of All Persons Index 2012=100 2
OPHPBS Business Sector: Real Output Per Hour of All Persons Index 2012=100 2
ULCBS Business Sector: Unit Labor Cost Index 2012=100 2
ULCMFG Manufacturing Sector: Unit Labor Cost Index 2012=100 2
ULCNFB Nonfarm Business Sector: Unit Labor Cost Index 2012=100 2
UNLPNBS Nonfarm Business Sector: Unit Nonlabor Payments Index 2012=100 2
UEMPMEAN Average (Mean) Duration of Unemployment Weeks 2
CES0600000007 Average Weekly Hours of Production and Nonsupervisory Employees: Goods-Producing Hours 2
CES0600000008 Average Hourly Earnings of Production and Nonsupervisory Employees: Goods-Producing Dollars per Hour 2

21
 CES0600000008 Average Hourly Earnings of Production and Nonsupervisory Employees: Goods-Producing Dollars per Hour 2
GDPC1 Real Gross Domestic Product Billions of Chained 2012 Dollars 3
GPDIC1 Real Gross Private Domestic Investment Billions of Chained 2012 Dollars 3
FPI Fixed Private Investment Billions of Dollars 3
Y033RC1Q027SBEA Gross Private Domestic Investment: Fixed Investment: Nonresidential: Equipment Billions of Dollars 3
PNFI Private Nonresidential Fixed Investment Billions of Dollars 3
PRFI Private Residential Fixed Investment Billions of Dollars 3
A014RE1Q156NBEA Shares of gross domestic product: Gross private domestic investment: Change in private inventories Percent 3
GCEC1 Real Government Consumption Expenditures and Gross Investment Billions of Chained 2012 Dollars 3
A823RL1Q225SBEA Real Government Consumption Expenditures and Gross Investment: Federal Percent Change from Preceding Period 3
FGRECPT Federal Government Current Receipts Billions of Dollars 3
SLCE State and Local Consumption Expenditures & Gross Investment Billions of Dollars 3
EXPGSC Real Exports of Goods and Services, 3 Decimal Billions of Chained 2012 Dollars 3
IMPGSC1 Real Imports of Goods and Services, 3 Decimal Billions of Chained 2012 Dollars 3
DPIC96 Real Disposable Personal Income Billions of Chained 2012 Dollars 3
GDPCTPI Gross Domestic Product: Chain-type Price Index Index 2012=100 3
GPDICTPI Gross Private Domestic Investment: Chain-type Price Index Index 2012=100 3
DGDSRG3Q086SBEA Personal consumption expenditures: Goods (chain-type price index) Index 2012=100 3
DDURRG3Q086SBEA Personal consumption expenditures: Durable goods (chain-type price index) Index 2012=100 3
DSERRG3Q086SBEA Personal consumption expenditures: Services (chain-type price index) Index 2012=100 3
DNDGRG3Q086SBEA Personal consumption expenditures: Nondurable goods (chain-type
Services: Household priceexpenditures
consumption index) (chain-type price Index 2012=100 3
DHCERG3Q086SBEA index) Index 2012=100 3
DMOTRG3Q086SBEA Motor vehicles
Personal consumption expenditures: Durable goods: Furnishings andand partshousehold
durable (chain-typeequipment
price index) Index 2012=100 3
DFDHRG3Q086SBEA (chain-typeconsumption
Personal price index)expenditures: Durable goods: Recreational goods and vehicles (chain-type Index 2012=100 3
DREQRG3Q086SBEA price index) Index 2012=100 3
DODGRG3Q086SBEA Personal consumption expenditures: Durable goods: Other durable goods (chain-type price index) Index 2012=100 3
Personal consumption expenditures: Nondurable goods: Food and beverages purchased for off-
DFXARG3Q086SBEA premises consumption expenditures:
Personal (chain-type price index) goods: Clothing and footwear (chain-type price
Nondurable Index 2012=100 3
DCLORG3Q086SBEA index)
Personal consumption expenditures: Nondurable goods: Gasoline and other energy goods (chain-type Index 2012=100 3
DGOERG3Q086SBEA price index)
Personal consumption expenditures: Nondurable goods: Other nondurable goods (chain-type price Index 2012=100 3
DONGRG3Q086SBEA index) Index 2012=100 3
DHUTRG3Q086SBEA Personal consumption expenditures: Services: Housing and utilities (chain-type price index) Index 2012=100 3
DHLCRG3Q086SBEA Personal consumption expenditures: Services: Health care (chain-type price index) Index 2012=100 3
DTRSRG3Q086SBEA Personal consumption expenditures: Transportation services (chain-type price index) Index 2012=100 3
DRCARG3Q086SBEA RecreationFood
Personal consumption expenditures: Services: services (chain-type
services price index) (chain-type price
and accommodations Index 2012=100 3
DFSARG3Q086SBEA index) Index 2012=100 3
DIFSRG3Q086SBEA Personal consumption expenditures: Financial services and insurance (chain-type price index) Index 2012=100 3
DOTSRG3Q086SBEA Personal consumption expenditures: Other services (chain-type price index) Index 2012=100 3
TABSHNO Households and Nonprofit Organizations; Total Assets, Level Billions of Dollars 3
TLBSHNO Households and Nonprofit Organizations; Total Liabilities, Level Billions of Dollars 3
CMDEBT Households and Nonprofit Organizations; Credit Market Instruments; Liability, Level Billions of Dollars 3
TNWBSHNO Households and Nonprofit Organizations; Net Worth, Level Billions of Dollars 3
TFAABSHNO Households and Nonprofit Organizations; Total Financial Assets, Level Billions of Dollars 3
HNOREMQ027S Households and nonprofit organizations; real estate at market value, Level Millions of Dollars 3
B020RE1Q156NBEA Shares of gross domestic product: Exports of goods and services Percent 3
B021RE1Q156NBEA Shares of gross domestic product: Imports of goods and services Percent 3
GFDEGDQ188S Federal Debt: Total Public Debt as Percent of Gross Domestic Product Percent of GDP 3
GFDEBTN Federal Debt: Total Public Debt Millions of Dollars 3
TLBSNNCB Nonfinancial Corporate Business; Total Liabilities, Level Billions of Dollars 3
TTAABSNNCB Nonfinancial Corporate Business; Nonfinancial Assets, Level Billions of Dollars 3
TNWMVBSNNCB Nonfinancial Corporate Business; Net Worth, Level Billions of Dollars 3
TLBSNNB Nonfinancial Corporate Business; Net Worth, Level Billions of Dollars 3
TABSNNB Nonfinancial noncorporate business; total assets, Level Millions of Dollars 3
TNWBSNNB Nonfinancial Noncorporate Business; Proprietors' Equity in Noncorporate Business (Net Worth), Level Billions of Dollars 3
CNCF Corporate Net Cash Flow with IVA Billions of Dollars 3 22
INDPRO Industrial Production Index Index 2012=100 4
IPFINAL Industrial Production: Final Products (Market Group) Index 2012=100 4
 ISRATIO Total Business: InventoriestoSalesRatio Ratio 4
PCECC96 Real Personal ConsumptionExpenditures Bil ionsofChained2012Dollars 5
PCDG Personal ConsumptionExpenditures: Durable Goods Bil ionsofDollars 5
PCESV Personal ConsumptionExpenditures: Services Bil ionsofDollars 5
PCND Personal ConsumptionExpenditures: Nondurable Goods Bil ionsofDollars 5
PCECTPI Personal ConsumptionExpenditures: Chain-typePriceIndex Index2012=100 5
PCEPILFE Personal ConsumptionExpendituresExcludingFoodandEnergy(Chain-TypePriceIndex) Index2012=100 5
CPIAUCSL ConsumerPriceIndexforAll UrbanConsumers: All Items Index2012=100 5
CPILFESL ConsumerPriceIndexforAll UrbanConsumers: All ItemsLessFoodandEnergy Index2012=100 5
WPSFD49207 ProducerPriceIndexbyCommodityforFinal Demand: FinishedGoods Index2012=100 5
PPIACO ProducerPriceIndexforAll Commodities Index2012=100 5
WPSFD49502 ProducerPriceIndexbyCommodityforFinal ConsumerGoods Index2012=100 5
WPSFD4111 ProducerPriceIndexbyCommodityforFinal Demand: FinishedConsumerFoods Index2012=100 5
PPIIDC ProducerPriceIndexbyCommodityIndustrial Commodities Index2012=100 5
WPUID6151 ProducerPriceIndexbyCommodityforIntermediateDemandbyCommodityType: SuppliesandComponents Index2012=100 5
WTISPLC SpotCrudeOil Price: WestTexasIntermediate(WTI) DollarsperBarrel 5
SPCS10RSA S&P/Case-Shil er10-CityCompositeHomePriceIndex Index2012=100 5
SPCS20RSA S&P/Case-Shil er20-CityCompositeHomePriceIndex© Index2012=100 5
WPSID62 ProducerPriceIndexbyCommodityforIntermediateDemandbyCommodityType: UnprocessedGoodsforIntermediateDemand Index2012=100 5
PPICMM ProducerPriceIndexbyCommodityMetals andmetal products: Primarynonferrousmetals Index2012=100 5
CPIAPPSL ConsumerPriceIndexforAll UrbanConsumers: Apparel Index2012=100 5
CPITRNSL ConsumerPriceIndexforAll UrbanConsumers: Transportation Index2012=100 5
CPIMEDSL ConsumerPriceIndexforAll UrbanConsumers: Medical Care Index2012=100 5
CUSR0000SAC ConsumerPriceIndexforAll UrbanConsumers: Commodities Index2012=100 5
CUUR0000SAD ConsumerPriceIndexforAll UrbanConsumers: Durables Index2012=100 5
CUSR0000SAS ConsumerPriceIndexforAll UrbanConsumers: Services Index2012=100 5
CPIULFSL ConsumerPriceIndexforAll UrbanConsumers: All ItemsLessFood Index2012=100 5
CUUR0000SA0L2 ConsumerPriceIndexforAll UrbanConsumers: All itemslessshelter Index2012=100 5
CUSR0000SA0L5 ConsumerPriceIndexforAll UrbanConsumers: All itemslessmedical care Index2012=100 5
CUSR0000SEHC ConsumerPriceIndexforAll UrbanConsumers: Owners' equivalentrentofresidences Index2012=100 5
NIKKEI225 Nikkei StockAverage, Nikkei 225 Index 6
USRECQ USRecession 1or0 7
RECPROUSM156N SmoothedU.S. RecessionProbabilities Percent 7
PERMITNE NewPrivateHousingUnitsAuthorizedbyBuildingPermitsin theNortheastCensusRegion ThousandsofUnits 7
PERMITMW NewPrivateHousingUnitsAuthorizedbyBuildingPermitsin theMidwestCensusRegion ThousandsofUnits 7
PERMITS NewPrivateHousingUnitsAuthorizedbyBuildingPermitsin theSouthCensusRegion ThousandsofUnits 7
PERMITW NewPrivateHousingUnitsAuthorizedbyBuildingPermitsin theWestCensusRegion ThousandsofUnits 7
USEPUINDXM Economic PolicyUncertaintyIndexforUnitedStates Index 8 23
 Appendix 2 – Data and Results
Group 1 Regression
Probit regression Number of obs = 91
LR chi2(14) = 58.72
Prob > chi2 = 0.0000
Log likelihood = -.00306055 Pseudo R2 = 0.9999

USRECQ Coef. Std. Err. z P>|z| [95% Conf. Interval]

FEDFUNDS 172.775 . . . . .
TB3MS 3.33764 . . . . .
TB6MS -171.1929 . . . . .
GS1 13.04458 . . . . .
GS10 7.58623 . . . . .
MORTGAGE30US 1.454506 . . . . .
AAA -9.429019 . . . . .
BAA 2.023008 . . . . .
BAA10YM 0 (omitted)
TB6SMFFM 188.3737 . . . . .
T1YFF -18.31383 . . . . .
T10YFFM 0 (omitted)
AMBSLREAL .0964378 . . . . .
IMFSL .0031731 1.938707 0.00 0.999 -3.796622 3.802969
M1REAL -.0399791 49.78006 -0.00 0.999 -97.6071 97.52714
M2REAL -1.062543 242.227 -0.00 0.997 -475.8187 473.6936
MZMREAL 1.529456 . . . . .
BUSLOANS -.0231453 4.499627 -0.01 0.996 -8.842252 8.795961
CONSUMERNSA .0059472 13.27982 0.00 1.000 -26.02203 26.03392
NONREVSL .1131075 23.38663 0.00 0.996 -45.72384 45.95005
REALLN -.0139162 2.837281 -0.00 0.996 -5.574884 5.547052
REVOLSL .1328804 . . . . .
TOTALSL -.1302241 11.766 -0.01 0.991 -23.19116 22.93071
DRIWCIL -.1785683 33.53444 -0.01 0.996 -65.90486 65.54772
EXSZUS -14.66663 . . . . .
EXJPUS -.0127739 31.39267 -0.00 1.000 -61.54127 61.51572
EXUSUK -2.111143 . . . . .
EXCAUS 13.0827 . . . . .
UMCSENT .0845779 32.98495 0.00 0.998 -64.56473 64.73388
TOTRESNS -.039075 5.221074 -0.01 0.994 -10.27219 10.19404
NONBORRES .000058 .0069186 0.01 0.993 -.0135021 .0136182
GS5 -18.38424 . . . . .
TB3SMFFM 0 (omitted)
T5YFFM 13.22707 . . . . .
AAAFFM 0 (omitted)
INVEST .0287787 5.563681 0.01 0.996 -10.87584 10.93339
DCPF3M 4.020571 . . . . .
NASDAQCOM -.0068348 14.26255 -0.00 1.000 -27.96093 27.94726
_cons -1.680813 . . . . .

Group 2 Regression
Probit regression Number of obs = 132
LR chi2(47) = 75.63
Prob > chi2 = 0.0051
Log likelihood = -.04820542 Pseudo R2 = 0.9987

USRECQ Coef. Std. Err. z P>|z| [95% Conf. Interval]

OUTNFB -121.7079 . . . . .
OUTBS 124.3366 507.9607 0.24 0.807 -871.2481 1119.921
OUTMS .367451 267.3628 0.00 0.999 -523.6541 524.389
PAYEMS -.0399503 . . . . .
USPRIV -.0007584 2.847896 -0.00 1.000 -5.582532 5.581015
MANEMP .043117 4.25557 0.01 0.992 -8.297648 8.383882
SRVPRD .0205759 1.941042 0.01 0.992 -3.783797 3.824949
USGOOD 0 (omitted)
DMANEMP -.0133483 3.486668 -0.00 0.997 -6.847092 6.820395
NDMANEMP 0 (omitted)
USCONS .0346951 2.42569 0.01 0.989 -4.719571 4.788961
USEHS .0071 3.191856 0.00 0.998 -6.248823 6.263023
USFIRE -.0006377 3.798098 -0.00 1.000 -7.444773 7.443498
USINFO .0229539 4.386785 0.01 0.996 -8.574986 8.620894
USPBS .0190323 3.033188 0.01 0.995 -5.925907 5.963972
USLAH .0172085 3.528835 0.00 0.996 -6.899182 6.933599
USSERV .0435877 3.214568 0.01 0.989 -6.25685 6.344025
USMINE 0 (omitted)
USTPU 0 (omitted)
USGOVT 0 (omitted)
USTRADE .0057965 3.011806 0.00 0.998 -5.897235 5.908828
USWTRADE .0315549 3.609471 0.01 0.993 -7.042878 7.105987
CES9091000001 .0030905 2.847262 0.00 0.999 -5.57744 5.583621
CES9092000001 .0112929 5.577946 0.00 0.998 -10.92128 10.94387
CES9093000001 0 (omitted)
CE16OV .003374 .2931723 0.01 0.991 -.5712332 .5779812
CIVPART -1.251355 676.5 -0.00 0.999 -1327.167 1324.664
UNRATE -26.29893 3676.885 -0.01 0.994 -7232.861 7180.263
LNS14000012 .3175136 212.5971 0.00 0.999 -416.3651 417.0001
LNS14000025 7.349437 1443.438 0.01 0.996 -2821.736 2836.435
LNS14000026 9.592444 977.5084 0.01 0.992 -1906.289 1925.474
UEMPLT5 .0095689 2.454152 0.00 0.997 -4.800481 4.819619
UEMP5TO14 -.0072812 1.295278 -0.01 0.996 -2.54598 2.531418
UEMP15T26 .0127908 3.070386 0.00 0.997 -6.005055 6.030637
UEMP27OV .0020437 2.04931 0.00 0.999 -4.014529 4.018617
LNS13023621 .0065784 1.143746 0.01 0.995 -2.235123 2.24828
LNS13023557 .017338 1.247587 0.01 0.989 -2.427888 2.462564
LNS13023705 .0108758 1.550227 0.01 0.994 -3.027513 3.049264
LNS13023569 -.0148424 2.225126 -0.01 0.995 -4.376009 4.346325
LNS12032194 -.0071223 .5292539 -0.01 0.989 -1.044441 1.030196
HOABS -111.0622 883.4271 -0.13 0.900 -1842.548 1620.423
HOAMS 1.00736 171.3392 0.01 0.995 -334.8112 336.8259
HOANBS 108.7162 . . . . .
AWHMAN 4.855341 1307.54 0.00 0.997 -2557.875 2567.586
AWHNONAG 6.932627 2134.574 0.00 0.997 -4176.756 4190.622
AWOTMAN -4.214278 691.9722 -0.01 0.995 -1360.455 1352.026
IPDBS 3.189161 . . . . .
CES2000000008 4.432692 767.9008 0.01 0.995 -1500.625 1509.491
CES3000000008 10.93254 2365.763 0.00 0.996 -4625.877 4647.743
COMPRMS 1.096524 316.1123 0.00 0.997 -618.4722 620.6652
COMPRNFB 53.50651 1171.45 0.05 0.964 -2242.493 2349.506
RCPHBS -53.76691 1018.606 -0.05 0.958 -2050.199 1942.665
OPHMFG .035044 10.28409 0.00 0.997 -20.1214 20.19148
OPHNFB 65.1401 722.8909 0.09 0.928 -1351.7 1481.98
OPHPBS -67.49648 . . . . .
ULCBS 50.61818 306.5577 0.17 0.869 -550.2239 651.4602
ULCMFG -.7327548 278.8143 -0.00 0.998 -547.1987 545.7332
ULCNFB -50.53122 . . . . .
UNLPNBS .0190249 103.2711 0.00 1.000 -202.3886 202.4267
CES0600000007 -7.159299 1055.445 -0.01 0.995 -2075.794 2061.476
CES0600000008 -10.57843 . . . . .
_cons 227.55 . . . . .

24
Group 3 Regression
Probit regression Number of obs = 213
LR chi2(47) = 161.95
Prob > chi2 = 0.0000
Log likelihood = -.00537694 Pseudo R2 = 0.9999

USRECQ Coef. Std. Err. z P>|z| [95% Conf. Interval]

GDPC1 .0676651 3.754778 0.02 0.986 -7.291565 7.426895

GPDIC1 -.167012 7.364156 -0.02 0.982 -14.60049 14.26647
FPI .4406038 8.849158 0.05 0.960 -16.90343 17.78463
Y033RC1Q027SBEA -.0898345 27.54253 -0.00 0.997 -54.07221 53.89254
PNFI -.3041197 23.3296 -0.01 0.990 -46.02929 45.42105
PRFI 0 (omitted)
A014RE1Q156NBEA 10.91206 230.5553 0.05 0.962 -440.9681 462.7922
GCEC1 -.0578364 6.423418 -0.01 0.993 -12.64751 12.53183
A823RL1Q225SBEA -.0837034 11.27134 -0.01 0.994 -22.17511 22.00771
FGRECPT -.0555747 3.875257 -0.01 0.989 -7.650939 7.53979
SLCE .0253264 18.16072 0.00 0.999 -35.56903 35.61968
EXPGSC .1965842 12.3299 0.02 0.987 -23.96957 24.36274
IMPGSC1 .1066112 8.964711 0.01 0.991 -17.4639 17.67712
DPIC96 -.0297867 1.806963 -0.02 0.987 -3.57137 3.511796
GDPCTPI 2.825888 931.8662 0.00 0.998 -1823.598 1829.25
GPDICTPI -.9337364 456.8555 -0.00 0.998 -896.3541 894.4866
DGDSRG3Q086SBEA 28.16882 885.9426 0.03 0.975 -1708.247 1764.584
DDURRG3Q086SBEA -29.52937 909.8605 -0.03 0.974 -1812.823 1753.764
DSERRG3Q086SBEA -124.5141 . . . . .
DNDGRG3Q086SBEA -16.0141 . . . . .
DHCERG3Q086SBEA 31.44478 . . . . .
DMOTRG3Q086SBEA 16.78227 547.9854 0.03 0.976 -1057.249 1090.814
DFDHRG3Q086SBEA 7.295128 221.6521 0.03 0.974 -427.1351 441.7254
DREQRG3Q086SBEA 2.961605 56.15834 0.05 0.958 -107.1067 113.0299
DODGRG3Q086SBEA 4.358708 175.3174 0.02 0.980 -339.2571 347.9745
DFXARG3Q086SBEA -4.923948 381.6256 -0.01 0.990 -752.8964 743.0485
DCLORG3Q086SBEA -.3413086 171.2798 -0.00 0.998 -336.0436 335.361
DGOERG3Q086SBEA -1.735407 174.468 -0.01 0.992 -343.6864 340.2156
DONGRG3Q086SBEA 1.764377 488.1422 0.00 0.997 -954.9767 958.5054
DHUTRG3Q086SBEA 41.61998 571.7691 0.07 0.942 -1079.027 1162.267
DHLCRG3Q086SBEA 7.976456 519.9482 0.02 0.988 -1011.103 1027.056
DTRSRG3Q086SBEA 1.778478 288.328 0.01 0.995 -563.334 566.891
DRCARG3Q086SBEA 3.603193 585.0884 0.01 0.995 -1143.149 1150.355
DFSARG3Q086SBEA 8.784245 495.554 0.02 0.986 -962.4837 980.0521
DIFSRG3Q086SBEA 8.148838 192.0378 0.04 0.966 -368.2384 384.536
DOTSRG3Q086SBEA 8.527354 357.5866 0.02 0.981 -692.3295 709.3842
TABSHNO -.0293307 9.631622 -0.00 0.998 -18.90696 18.8483
TLBSHNO -.0063103 9.246664 -0.00 0.999 -18.12944 18.11682
CMDEBT .018811 8.455882 0.00 0.998 -16.55441 16.59204
TNWBSHNO 0 (omitted)
TFAABSHNO .0304401 9.613782 0.00 0.997 -18.81223 18.87311
HNOREMQ027S .0000397 .0112983 0.00 0.997 -.0221046 .022184
B020RE1Q156NBEA 1.492844 688.3926 0.00 0.998 -1347.732 1350.717
B021RE1Q156NBEA -8.419559 462.8774 -0.02 0.985 -915.6426 898.8034
GFDEGDQ188S 1.603267 126.0251 0.01 0.990 -245.4014 248.608
GFDEBTN -1.23e-06 .0018534 -0.00 0.999 -.0036338 .0036313
TLBSNNCB -.0083074 .6931204 -0.01 0.990 -1.366798 1.350184
TTAABSNNCB .0307341 2.113647 0.01 0.988 -4.111938 4.173406
TNWMVBSNNCB .004924 .6804463 0.01 0.994 -1.328726 1.338574
TLBSNNB .0364924 4.449211 0.01 0.993 -8.683801 8.756785
TABSNNB -.0999085 6.334459 -0.02 0.987 -12.51522 12.3154
TNWBSNNB 0 (omitted)
CNCF -.0398318 2.856903 -0.01 0.989 -5.639259 5.559595
_cons -411.4353 8522.365 -0.05 0.961 -17114.96 16292.09

Group 4 Regression
Probit regression Number of obs = 112
LR chi2(25) = 62.63
Prob > chi2 = 0.0000
Log likelihood = -.00644019 Pseudo R2 = 0.9998

USRECQ Coef. Std. Err. z P>|z| [95% Conf. Interval]

INDPRO -22.14525 . . . . .
IPFINAL 1.711152 334.9001 0.01 0.996 -654.681 658.1033
IPCONGD -4.077012 870.8404 -0.00 0.996 -1710.893 1702.739
IPMAT 11.75663 62.10472 0.19 0.850 -109.9664 133.4796
IPDMAT -1.066738 88.69075 -0.01 0.990 -174.8974 172.7639
IPNMAT -.3485021 77.03285 -0.00 0.996 -151.3301 150.6331
IPDCONGD .4201292 324.094 0.00 0.999 -634.7925 635.6327
IPB51110SQ .3702368 90.12724 0.00 0.997 -176.2759 177.0164
IPNCONGD 4.809602 504.043 0.01 0.992 -983.0965 992.7157
IPBUSEQ .5415214 77.71384 0.01 0.994 -151.7748 152.8578
IPB51220SQ 6.524268 282.7074 0.02 0.982 -547.572 560.6205
TCU -2.643053 700.1681 -0.00 0.997 -1374.947 1369.661
CUMFNS 3.084976 586.842 0.01 0.996 -1147.104 1153.274
HOUST -.0249838 2.996186 -0.01 0.993 -5.8974 5.847433
HOUST5F -.0325545 3.439121 -0.01 0.992 -6.773108 6.707999
PERMIT .0152949 1.30549 0.01 0.991 -2.543418 2.574008
HOUSTMW .0441122 5.725853 0.01 0.994 -11.17835 11.26658
HOUSTNE -.0160363 10.87364 -0.00 0.999 -21.32798 21.29591
HOUSTS .0378149 5.649568 0.01 0.995 -11.03513 11.11076
HOUSTW 0 (omitted)
CMRMTSPL .0001712 .0146505 0.01 0.991 -.0285433 .0288857
RSAFS -.000471 .0369871 -0.01 0.990 -.0729644 .0720223
INVCQRMTSPL -.0000186 .0033903 -0.01 0.996 -.0066635 .0066263
USSTHPI .2982812 28.55255 0.01 0.992 -55.66368 56.26025
IPMANSICS 7.437093 340.9028 0.02 0.983 -660.7201 675.5943
IPB51222S -3.526788 164.4385 -0.02 0.983 -325.8203 318.7667
IPFUELS -2.601996 171.8471 -0.02 0.988 -339.4161 334.2121
GF 0 (omitted)
GG 0 (omitted)
_cons -29.44274 16568.41 -0.00 0.999 -32502.93 32444.04

25
Group 5 Regression
Probit regression Number of obs = 132
LR chi2(10) = 75.72
Prob > chi2 = 0.0000
Log likelihood = -.00001107 Pseudo R2 = 1.0000

USRECQ Coef. Std. Err. z P>|z| [95% Conf. Interval]

PCECC96 -.0335862 49.30091 -0.00 0.999 -96.66159 96.59442

PCDG -.0788489 301.8025 -0.00 1.000 -591.6009 591.4432
PCESV -.1245894 52.65228 -0.00 0.998 -103.3212 103.072
PCND .6876653 138.1159 0.00 0.996 -270.0145 271.3898
PCECTPI -144.8575 . . . . .
PCEPILFE 128.6982 . . . . .
CPIAUCSL -123.6909 . . . . .
CPILFESL -52.63145 . . . . .
WPSFD49207 -17.26499 . . . . .
PPIACO 4.422468 . . . . .
WPSFD49502 7.14431 . . . . .
WPSFD4111 -2.172753 921.8845 -0.00 0.998 -1809.033 1804.688
PPIIDC 3.750383 . . . . .
WPUID6151 -5.436033 . . . . .
WTISPLC -.1153588 540.4711 -0.00 1.000 -1059.419 1059.188
SPCS10RSA -3.674006 1928.368 -0.00 0.998 -3783.206 3775.858
SPCS20RSA 1.065438 1770.82 0.00 1.000 -3469.678 3471.808
WPSID62 .5375309 2886.314 0.00 1.000 -5656.533 5657.608
PPICMM -.1920761 2148.026 -0.00 1.000 -4210.245 4209.861
CPIAPPSL .6750738 839.1503 0.00 0.999 -1644.029 1645.379
CPITRNSL .641306 . . . . .
CPIMEDSL -4.730801 . . . . .
CUSR0000SAC 157.7323 . . . . .
CUUR0000SAD 2.909456 . . . . .
CUSR0000SAS 244.9325 . . . . .
CPIULFSL -60.37557 . . . . .
CUUR0000SA0L2 -18.53459 . . . . .
CUSR0000SA0L5 -124.4042 . . . . .
_cons 387.1515 . . . . .

Group 6 Regression
Probit regression Number of obs = 260
LR chi2(1) = 8.05
Prob > chi2 = 0.0045
Log likelihood = -94.90322 Pseudo R2 = 0.0407

USRECQ Coef. Std. Err. z P>|z| [95% Conf. Interval]

NIKKEI225 .0000378 .000014 2.69 0.007 .0000103 .0000653

_cons .7897219 .1558542 5.07 0.000 .4842533 1.09519

Group 7 Regression
Probit regression Number of obs = 240
LR chi2(1) = 23.27
Prob > chi2 = 0.0000
Log likelihood = -78.790627 Pseudo R2 = 0.1287

USRECQ Coef. Std. Err. z P>|z| [95% Conf. Interval]

HN .0014978 .0003405 4.40 0.000 .0008304 .0021652

_cons -.7052385 .4146957 -1.70 0.089 -1.518027 .10755

Group 8 Regression
Probit regression Number of obs = 140
LR chi2(1) = 8.03
Prob > chi2 = 0.0046
Log likelihood = -34.520924 Pseudo R2 = 0.1042

USRECQ Coef. Std. Err. z P>|z| [95% Conf. Interval]

USEPUINDXM -.015215 .00551 -2.76 0.006 -.0260144 -.0044156

_cons 3.219094 .7050436 4.57 0.000 1.837234 4.600954

26
Large Model Regression
Probit regression Number of obs = 132
LR chi2(47) = 75.63
Prob > chi2 = 0.0051
Log likelihood = -.04820542 Pseudo R2 = 0.9987

USRECQ Coef. Std. Err. z P>|z| [95% Conf. Interval]

OUTNFB -121.7079 . . . . .
OUTBS 124.3366 507.9607 0.24 0.807 -871.2481 1119.921
OUTMS .367451 267.3628 0.00 0.999 -523.6541 524.389
PAYEMS -.0399503 . . . . .
USPRIV -.0007584 2.847896 -0.00 1.000 -5.582532 5.581015
MANEMP .043117 4.25557 0.01 0.992 -8.297648 8.383882
SRVPRD .0205759 1.941042 0.01 0.992 -3.783797 3.824949
USGOOD 0 (omitted)
DMANEMP -.0133483 3.486668 -0.00 0.997 -6.847092 6.820395
NDMANEMP 0 (omitted)
USCONS .0346951 2.42569 0.01 0.989 -4.719571 4.788961
USEHS .0071 3.191856 0.00 0.998 -6.248823 6.263023
USFIRE -.0006377 3.798098 -0.00 1.000 -7.444773 7.443498
USINFO .0229539 4.386785 0.01 0.996 -8.574986 8.620894
USPBS .0190323 3.033188 0.01 0.995 -5.925907 5.963972
USLAH .0172085 3.528835 0.00 0.996 -6.899182 6.933599
USSERV .0435877 3.214568 0.01 0.989 -6.25685 6.344025
USMINE 0 (omitted)
USTPU 0 (omitted)
USGOVT 0 (omitted)
USTRADE .0057965 3.011806 0.00 0.998 -5.897235 5.908828
USWTRADE .0315549 3.609471 0.01 0.993 -7.042878 7.105987
CES9091000001 .0030905 2.847262 0.00 0.999 -5.57744 5.583621
CES9092000001 .0112929 5.577946 0.00 0.998 -10.92128 10.94387
CES9093000001 0 (omitted)
CE16OV .003374 .2931723 0.01 0.991 -.5712332 .5779812
CIVPART -1.251355 676.5 -0.00 0.999 -1327.167 1324.664
UNRATE -26.29893 3676.885 -0.01 0.994 -7232.861 7180.263
LNS14000012 .3175136 212.5971 0.00 0.999 -416.3651 417.0001
LNS14000025 7.349437 1443.438 0.01 0.996 -2821.736 2836.435
LNS14000026 9.592444 977.5084 0.01 0.992 -1906.289 1925.474
UEMPLT5 .0095689 2.454152 0.00 0.997 -4.800481 4.819619
UEMP5TO14 -.0072812 1.295278 -0.01 0.996 -2.54598 2.531418
UEMP15T26 .0127908 3.070386 0.00 0.997 -6.005055 6.030637
UEMP27OV .0020437 2.04931 0.00 0.999 -4.014529 4.018617
LNS13023621 .0065784 1.143746 0.01 0.995 -2.235123 2.24828
LNS13023557 .017338 1.247587 0.01 0.989 -2.427888 2.462564
LNS13023705 .0108758 1.550227 0.01 0.994 -3.027513 3.049264
LNS13023569 -.0148424 2.225126 -0.01 0.995 -4.376009 4.346325
LNS12032194 -.0071223 .5292539 -0.01 0.989 -1.044441 1.030196
HOABS -111.0622 883.4271 -0.13 0.900 -1842.548 1620.423
HOAMS 1.00736 171.3392 0.01 0.995 -334.8112 336.8259
HOANBS 108.7162 . . . . .
AWHMAN 4.855341 1307.54 0.00 0.997 -2557.875 2567.586
AWHNONAG 6.932627 2134.574 0.00 0.997 -4176.756 4190.622
AWOTMAN -4.214278 691.9722 -0.01 0.995 -1360.455 1352.026
IPDBS 3.189161 . . . . .
CES2000000008 4.432692 767.9008 0.01 0.995 -1500.625 1509.491
CES3000000008 10.93254 2365.763 0.00 0.996 -4625.877 4647.743
COMPRMS 1.096524 316.1123 0.00 0.997 -618.4722 620.6652
COMPRNFB 53.50651 1171.45 0.05 0.964 -2242.493 2349.506
RCPHBS -53.76691 1018.606 -0.05 0.958 -2050.199 1942.665
OPHMFG .035044 10.28409 0.00 0.997 -20.1214 20.19148
OPHNFB 65.1401 722.8909 0.09 0.928 -1351.7 1481.98
OPHPBS -67.49648 . . . . .
ULCBS 50.61818 306.5577 0.17 0.869 -550.2239 651.4602
ULCMFG -.7327548 278.8143 -0.00 0.998 -547.1987 545.7332
ULCNFB -50.53122 . . . . .
UNLPNBS .0190249 103.2711 0.00 1.000 -202.3886 202.4267
CES0600000007 -7.159299 1055.445 -0.01 0.995 -2075.794 2061.476
CES0600000008 -10.57843 . . . . .
_cons 227.55 . . . . .

27
Appendix 3 - Recession Regression Model Code

28
29
30
31
32
33
34
 References

Albrecht Ritschl, Samad Sarferaz and Martin Uebele, 2016. The U.S. Business Cycle, 1867–2006: A
Dynamic Factor Approach. [online] [Link]. Available at:
<[Link] [Accessed 30 April 2020].
Alessia Paccagnini, 2017. Forecasting With FAVAR: Macroeconomic Versus Financial Factors. [online]
[Link]. Available at: <[Link]
[Link]> [Accessed 30 April 2020].
Bai and NG, 2002. Determining The Number Of Factors In Approximate Factor Models. [online]
[Link]. Available at: <[Link] [Accessed 30
April 2020].
Bai and NG, 2004. A PANIC Attack On Unit Roots And Cointegration. [online] [Link]. Available at:
<[Link]
> [Accessed 30 April 2020].
Chen, Z., Iqbal and Lai, 2011. Forecasting The Probability Of US Recessions: A Probit And Dynamic
Factor Modelling Approach. [online] [Link]. Available at:
<[Link] [Accessed 30 April 2020].
Christiansen et al., 2014. Forecasting US Recessions: The Role Of Sentiments. [online]
[Link]. Available at:
<[Link] [Accessed 30
April 2020].
Christine De Mol, Domenico Giannone and Lucrezia Reichlin, 2006. FORECASTING USING A LARGE
NUMBER OF PREDICTORS - IS BAYESIAN REGRESSION A VALID ALTERNATIVE TO PRINCIPAL
COMPONENTS?. [online] [Link]. Available at:
<[Link] [Accessed 30 April 2020].
Duo Qin, 2008. Rise Of VAR Modelling Approach. [online] SSRN. Available at:
<[Link] [Accessed 30 April 2020].
Duo Qin, 2010. Https://[Link]/Sef/Media/Econ/Research/Workingpapers/Archive/[Link].
[online] [Link]. Available at:
<[Link] [Accessed 30
April 2020].
Qin, D., 2010. Modelling Of The Inflation-Unemployment Tradeoff From The Perspective Of The History Of
Econometrics. [online] [Link]. Available at:
<[Link] [Accessed 30 April 2020].
Duo Qin, 2013. A History Of Econometrics. [online] Google Books. Available at:
<[Link]
ary_r&cad=0#v=onepage&q&f=false> [Accessed 30 April 2020].
Fabio Fornari and Wolfgang Lemke, 2010. PREDICTING RECESSION PROBABILITIES WITH
FINANCIAL VARIABLES OVER MULTIPLE HORIZONS. [online] [Link]. Available at:
<[Link] [Accessed 30 April 2020].
Finn E. Kydland and Edward C. Prescott, 1982. Time To Build And Aggregate Fluctuations. [online] JStor.
Available at: <[Link] [Accessed 30 April 2020].
Giannone, Reichlin and Sala, 2004. Monetary Policy In Real Time. [online] [Link]. Available
at: <[Link] [Accessed 30 April
2020].

35
Hamilton, J., 1989. A New Approach To The Economic Analysis Of Nonstationary Time Series And The
Business Cycle. [online] [Link]. Available at:
<[Link]
[Accessed 30 April 2020].
James H. Stock and Mark W. Watson, 1993. A Procedure For Predicting Recessions With Leading
Indicators: Econometric Issues And Recent Experience. [online] [Link]. Available at:
<[Link]
[Accessed 30 April 2020].
James H. STOCK and Mark W. WATSON, 2002. Forecasting Using Principal Components From A Large
Number Of Predictors. [online] [Link]. Available at:
<[Link] [Accessed 30 April
2020].
James. H. Stock and Mark W. Watson, 2006. Forecasting And Now-Casting With Disparate Predictors:
Dynamic Factor Models And Beyond. [online] [Link]. Available at:
<[Link] [Accessed 30 April 2020].
Karim Barhoumi, Olivier Darné and Laurent Ferrara, 2014. Dynamic Factor Models: A Review Of The
Literature. [online] Research Gate. Available at:
<[Link] [Accessed 30 April 2020].
Kenneth Train, 2020. Discrete Choice Methods With Simulation. [online] Google Books. Available at:
<[Link]
[Accessed 30 April 2020].
Mark W. Watson, 1991. Using Econometric Models To Predict Recessions. [online] [Link].
Available at: <[Link] [Accessed 30 April
2020].
McMillan, B., 2019. The Yield Curve Has Un-Inverted: Now What?. [online] Forbes. Available at:
<[Link]
[Accessed 30 April 2020].
Nyberg, H., 2010. Dynamic Probit Models And Financial Variables In Recession Forecasting. [online]
Wiley. Available at: <[Link] [Accessed 30 April
2020].
Quah, D. and Sargent, 1992. A Dynamic Index Model For Large Cross Sections. [online] [Link].
Available at: <[Link] [Accessed 30 April 2020].
Salih [Link]çi, 2008. Optimal Prediction Of Cyclical Downturns. [online] Science Direct. Available at:
<[Link] [Accessed 30 April 2020].
Sargent, T. and Sims, C., 1977. Business Cycle Modeling Without Pretending To Have Too Much A Priori
Economic Theory. [online] [Link]. Available at:
<[Link] [Accessed 30 April 2020].
Serena Ng and Jonathan H. Wright, 2013. Facts And Challenges From The Great Recession For
Forecasting And Macroeconomic Modeling. [online] [Link]. Available at:
<[Link] [Accessed 30 April 2020].
[Link]. n.d. Dynamic-Factor Models. [online] Available at:
<[Link] [Accessed 30 April 2020].
James H. Stock and Mark W. Watson, 1989. New Indexes Of Coincident And Leading Economic Indicators.
[online] [Link]. Available at: <[Link] [Accessed 30 April
2020].
36
Stock, J. and Watson, M., 1992. A Procedure For Predicting Recessions With Leading Indicators:
Econometric Issues And Recent Experience. [online] National Bureau of Economic Research. Available
at: <[Link] [Accessed 30 April 2020].
Stock, J. and Watson, M., 2006. Forecasting With Many Predictors. [online] [Link].
Available at: <[Link] [Accessed 30 April 2020].
Tjalling C. Koopmans, 1949. The Econometric Approach To Business Fluctuations. [online] Jstor. Available
at: <[Link] [Accessed 30 April 2020].
Tinbergen, J., 1939. Statistical Testing Of Business Cycle Theories: Part I: A Method And Its Application To
Investment Activity. [online] [Link]. Available at: <[Link] [Accessed 30
April 2020].
Thomas Hsu, 2016. U.S. Recession Forecasting Using Probit Models With Asset Index Predictor Variables.
[online] [Link]. Available at: <[Link]
[Link]> [Accessed 30 April 2020].
Watson, M. and Stock, J., 1993. Business Cycles, Indicators, And Forecasting. Chicago, Ill.: University of
Chicago Press.
Wink, B., 2019. The Yield Curve Is Inverted. Here's What That Means, And What The Implications Are For
The Economy. | Markets Insider. [online] [Link]. Available at:
<[Link]
means-2019-8-1028482016#what-is-the-yield-curve-1> [Accessed 30 April 2020].

37