Vol. 11, 2017-9 | April 13, 2017 | http://dx.doi.org/10.5018/economics-ejournal.ja.

2017-9

Global shocks and their impact on the Tanzanian

economy
Fiseha Haile

Abstract
Plummeting commodity prices, China’s economic slowdown and rebalancing, and global
financial market turbulence have recently raised concerns about their effects on African
economies. This paper investigates whether, and to what extent, these intertwined shocks
spillover into the Tanzanian economy. The author finds that a 1 percentage point (ppts)
drop in China’s investment growth is associated with a decline in Tanzania’s export growth
of roughly 0.60 ppts. A 1 percent fall in commodity prices leads to 0.65 percent lower
exports value. The results suggest that a hard landing of the Chinese economy to its ‘new
normal’ would doubtless send shock waves through the Tanzanian economy by further
driving down commodity demand and prices as well as lowering development finance.
In contrast, financial market volatility has a fairly negligible impact on economic growth.
The main results stand up well to a wide-array of robustness checks.
JEL C32 F4 O11
Keywords China; Tanzania; commodity prices; investment; exports; Cointegrated
VAR model
Authors
Fiseha Haile, World Bank, Addis Ababa, Ethiopia, [email protected]

Citation Fiseha Haile (2017). Global shocks and their impact on the Tanzanian
economy. Economics: The Open-Access, Open-Assessment E-Journal, 11 (2017-9): 1—
38. http://dx.doi.org/10.5018/economics-ejournal.ja.2017-9

Received November 11, 2016 Published as Economics Discussion Paper November 24, 2016
Revised March 11, 2017 Accepted March 21, 2017 Published April 13, 2017
© Author(s) 2017. Licensed under the Creative Commons License - Attribution 4.0 International (CC BY 4.0)
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

1 Introduction
Tanzania has posted high and sustained economic growth over the past decade,
hovering around 6−7 percent. In addition, in recent years, inflation has been tamed
to reasonable single digit levels. Tanzania has also maintained a broadly stable
current account deficit. However, notwithstanding the overall positive short- to
medium-term outlook (World Bank, 2016), the economy is not fully resilient to
externally-induced shocks and, like many African economies, has recently been
facing several growing and intertwined risks, including China’s economic
slowdown, falling global commodity prices and, to a lesser extent, increased
volatility in global financial and foreign exchange markets.
Over the past decade or so, trade and investment links between Tanzania and
China have reached historically unprecedented levels. China is now one of
Tanzania’s biggest trading partners and increasingly important source of
development finance. However, although the increased economic ties are likely to
have bolstered economic growth, they have doubtless increased Tanzania’s
vulnerability to the vagaries of the Chinese economy. China’s investment-propelled
growth seems to be running out of steam, partly reflecting rebalancing towards a
consumption-driven and services-oriented economy (Lakatos et al., 2016). 1 The
structural shift has recently manifested itself in flagging demand and prices for
commodities. China’s faltering growth may engender a significant knock-on effect
on the Tanzanian economy via depressed export growth and potentially lower
development finance.
Plummeting commodity prices also pose risks to Tanzania by virtue of its
position as a primary exporter. Global commodity prices have generally been on a
downward spiral mainly on account of falling demand in China and higher
production capacity. The prices of Tanzania’s major export commodities, notably
gold, are at record lows, despite a slight resurgence in recent months. This
constituted the main factor underlying the country’s worsening terms-of-trade
during the past few years. Falling oil prices have partly dampened the deterioration
in the external balance as the country is a net importer of oil. However, the
implications of soft commodity prices need to be carefully assessed given that a

1
The Chinese economy has been undergoing significant changes. According to its 12th Five Year
Plan, China aims to rebalance the economy away from investment to consumption (see Lakatos et al.,
2016). In particular, the following policy measures, among others, are expected to facilitate China’s
structural transformation: (i) policies that promote urbanization and boost domestic consumption; and
(ii) policies that encourage lower domestic savings, which would result in lower investment. One of
the most important transmission channels through which these changes would affect Tanzania
constitutes trade. Specifically, China’s economic rebalancing would likely lower external demand for
Tanzanian exports.

www.economics-ejournal.org 2
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

wrong combination of price fluctuations (for instance, a continued decline in gold

prices and rebounding oil prices) might put a dent in the country’s respectable
growth.
Another cause for concern has been increased volatility in global financial and
foreign exchange markets. Tanzania remained largely unscathed by previous
financial market turbulences due to its limited financial development and global
integration. However, since the country is drifting towards deeper financial
integration, with rising private capital flows and external commercial borrowing as
well as pending sovereign bond issuance, it has become increasingly prone to
global market instabilities. A closer scrutiny thus seems warranted in light of
surfacing concerns that increased global financial volatility might put a drag on
Tanzania’s growth pace. Further, since the early 2015, the Tanzanian Shilling has
seen significant depreciation on the back of a strong dollar appreciation and, to a
limited extent, declining aid inflows. Hence the need for examining whether the
sharp nominal depreciation has been associated with higher inflation.
The present paper is an attempt to explore whether, and to what degree, the
aforementioned economic shocks spillover into the Tanzanian economy. Towards
this end, we employ the Cointegrated VAR model as a statistical benchmark. The
empirical estimates suggest that a 1 percentage point (ppts) decline in China’s
investment growth is associated with 0.57 ppts decrease in Tanzania’s export
growth. This underscores the importance of diversifying markets destination to
mitigate headwinds from demand fluctuations. In addition, a 1 percent lower export
commodity prices leads to a 0.65 percent decline in exports value, reflecting the
fact that Tanzanian exports are predominated by less diversified and largely
unprocessed primary commodities, and thus significantly prone to turbulences in
commodity prices. Moreover, a 1 ppts increase in capital flow volatility would
reduce economic growth by a negligible 0.01 ppts. Finally, the impact of a 1
percent depreciation of the nominal effective exchange rate is to increase the
inflation rate by around 0.58 ppts, albeit offset by the inflation-reducing impacts of
low oil and food prices.
In a nutshell, the paper deals with the following main questions: What are the
major external risks facing the Tanzanian economy and what are the main
transmission mechanisms through which they operate? Does a slowdown in
China’s investment-driven growth have a significant negative impact on
Tanzania’s export performance? How large is the spillover from falling global
commodity prices to the Tanzanian economy? What is the contribution of private
capital flows to the national economy and how does its volatility impact on
economic growth? Has the steep depreciation of the Shilling increased domestic
prices and helped fuel inflation?

www.economics-ejournal.org 3
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

The remainder of the paper is organized as follows: Section 2 discusses the

main sources of external risks to the Tanzanian economy? Section 3 briefly
presents the theoretical framework. Section 4 discusses the data, while Section 5 is
devoted to empirical model specification. Section 6 discusses the empirical results.
Finally, Section 7 winds up with concluding remarks.

2 External risks to the Tanzanian economy: An overview

Like many SSA economies, Tanzania has established unprecedented trade and
investment links with China over the past decade or so; hence more vulnerable to
China’s business cycle. In 2014, total Sino-Tanzania trade surged to about $2.6
billion from negligible levels in 2000. China is the third major export destination,
absorbing about 13 percent of the country’s total exports. The bulk of Tanzania’s
exports to China is accounted for by mineral and precious metal exports. The
country’s exports plunged in the early 2010s due to a soft demand in China,
although it somewhat rebounded in 2014 (Figure 1). Over the last decade, China’s
unparalleled growth was mainly propped up by an investment boom, which in turn
gave rise to soaring import demand for commodities and hence their prices.
China’s shift from commodity-intensive investment-led growth to services-driven
economy has thus depressed commodity demand, notably those for metals.2
Therefore, a contraction in exports is the primary transmission channel through
which Tanzania may feel the impact of a slowdown in the Chinese economy.
A sharper-than-expected slowdown in China could also affect the Tanzanian
economy via potentially lower development finance and FDI. China’s development
loans to Tanzania have grown substantially over the past decade and have been
instrumental in addressing the country’s severe infrastructure deficits. In the short-
term, economic downturn in China seems less likely to have a dramatic impact on
its economic engagement in Tanzania. However, the magnitude of the ripple effect
in the future depends on how successful and smooth China’s economic rebalancing
will be. To be sure, a hard landing of China’s economy to its ‘new normal’ would
have a considerable negative impact on the Tanzanian economy via causing
cutbacks in development finance.
The officially reported stock of Chinese FDI in Tanzania has increased more
than six fold and was roughly estimated at $70 million in 2014, albeit still small in

2
China saw a sharp increase in its share of global consumption of metals, from below 20 percent in
the early 2000s to more than 50 percent currently (IMF, 2015).

www.economics-ejournal.org 4
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Figure 1. Tanzania's Exports to Figure 2. Tanzania: Export Price

China (2000−2014), Mil. USD Index (2000=100)
300
600
250

400 200

150
200
100
0
50
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014

1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
2014
Gross exports
Source: IMF.
Source: UN Comtrade.

Figure 3. Net FDI Inflows Figure 4. Nominal Effective

(1990−2014), Mil. USD Exchange Rate (Jan. 2014− Oct. 2015)
97
2000
96
95
1500
94
1000 93
92
500 91
90
0
Jul 14

Jul 15
Jan 14
Mrz 14
Mai 14

Jan 15
Mrz 15
Mai 15
Sep 14
Nov 14

Sep 15
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
2014

Source: World Bank. Source: IMF.

absolute terms. If China’s economic growth loses it vigor, it may also put further
strain on the Tanzanian economy through possible deceleration in FDI inflows.
Tightening financial conditions in China could slow Chinese firms’ investments in
Tanzania as they now draw funds from a shrinking pool. However, fluctuations in
Chinese FDI are unlikely to send shockwaves through the economy as it makes up
less than 1 percent of the total FDI stock. In addition, as China moves away from
an investment-driven growth model, its state-backed firms may look for profitable
opportunities abroad and Africa may be a prime destination considering that
returns are large, competition is relatively limited, and the availability of cheap and
plentiful labor.
The downward spiral in global commodity prices is another source of concern,
which has been partly reinforced by China’s slowing growth trajectory. Despite the
demand-driven surge in global commodity prices that started around 2002 and
lasted more than a decade, often dubbed as the commodity “super-cycle”, prices

www.economics-ejournal.org 5
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

dropped steeply in recent years (Figure 2). The sharp slide in commodity prices has
been instigated by weaker demand and higher production capacity. Oil prices
witnessed a particularly large drop, reflecting resilient supply. Metal prices have
also plummeted. Gold price, for instance, stood at $1,106 per ounce in March 2016,
plunging nearly 75 percent from its peak in mid-2011. Tanzania’s gold exports fell
dramatically from their peak of $2.3 billion in 2011/12 to around $1.3 billion in
2014/15. As Tanzania is a net oil importer, markedly lower oil prices have relieved
its energy import bill. In fact, on the whole, the turbulence in commodity prices has
not hitherto severely affected Tanzania. However, an unfortunate turn of events
entailing, for instance, a significant drop in gold prices and a rebound in oil prices,
would likely have an adverse impact on the country’s growth performance.
Low commodity prices might also negatively affect future outputs as their
value is marked up or down with price changes. Tanzania has discovered vast
reserves of natural gas and massive foreign investment is expected to flow to its
near-shore gas sector. However, the persistent drop in the prices of oil and
liquefied natural gas (LNG) could weigh on the sentiment of multinational
companies and stifle investment in the short-term. Similarly, a drastic decline in
mineral prices might lead to a scaling down of existing and new operations in the
medium- to long-term. 3
In addition, Tanzania may be affected by rising volatility in global financial
markets. The recent bout of volatility spiked in the aftermath of heightened global
risk aversion associated with the financial turbulence in China and somewhat
bleaker prospects for emerging economies in general. Tanzania remained broadly
unaffected by previous financial market turbulences due to its limited financial
development and integration. However, as the county is gravitating towards deeper
financial integration, with rising private capital flows and external commercial
borrowing as well as pending sovereign bond issuance, it has become increasingly
susceptible to global market instabilities. Net FDI inflows to Tanzania stood at $2.1
billion in 2014, from less than $400 million a decade ago (Figure 3), reflecting the
country’s increasing reliance on potentially volatile private capital flows. 4
Increased volatility in financial markets can affect Tanzania mainly by causing
disruptions to capital flows. 5 However, with FDI flows to Africa projected to

3
However, setting up investments in mining is expensive and hence operations that are underway
would likely be reversed in the short- to medium-term only if prices fell short of variable costs. This,
however, seems likely only in some mature investments.
4
In contrast, portfolio flows to Tanzania account for only a meager proportion of total private capital
flows, testifying to the underdevelopment of the domestic capital market and restrictions on capital
account transactions.
5
As the country’s domestic bond market does not rely on foreign investors, there should be no risk of
portfolio reallocation by nonresident investors.

www.economics-ejournal.org 6
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

remain stable in the short-term (World Bank, 2015a), the impact of volatility in
financial markets may be modest. If global volatility intensifies, however, the
country could suffer from a slowdown or possible reversal of capital flows and
uncertainties over future aid inflows. In recent years, low interest rates in
international markets and subdued financial volatility have led a number of SSA
economies to issue sovereign bonds. However, in the face of increasing global
volatilities and US interest rates, Tanzania’s future plans to issue bonds are likely
to entail higher costs.
Turning to developments in the foreign exchange market, the Shilling
depreciated sharply against the currencies of Tanzania’s major trading partners,
except the Euro. As of January 2015, the nominal depreciation against the US
dollar stood at nearly 30 percent year-on-year (see Figure 4). The large
depreciation was mainly on account of strong dollar appreciation and, to a lesser
extent, a decline in aid inflows. Despite the substantial Shilling depreciation, the
inflation rate has been fairly stable and remained within single-digit levels.
However, this does not necessarily imply that the former had no impact on
inflation. The reason is that low and stable inflation could be the result of a
confluence of counteracting factors. For example, falling oil prices might have
neutralized the inflationary impact of currency depreciation. This necessitates
untangling the impact of exchange rate movements on inflation.

3 Theoretical framework
3.1 China’s slowdown, commodity prices, and Tanzania’s exports
We are generally interested in addressing the central question of whether, and the
extent to which, Tanzania is susceptible to China’s economic slowdown. In
particular, the analysis attempts to quantitatively pin down trade spillovers from
China’s potentially slower, and more balanced, growth into the Tanzanian
economy. The impact of changes in China’s domestic investment on Tanzanian
exports is examined based on a model that takes the following form6 (hereafter
referred to as Model 1):

exportt = 𝜃0 + 𝜃1 cdit + 𝜃2 pricet + 𝜃3 yworld

t
+ 𝜀𝑡 (1)
∆exportt = 𝜃0∗ + 𝜃1∗ ∆cdit + 𝜃2∗ ∆pricet + 𝜃3∗ ∆yworld
t
(2)

6
This specification draws mainly on Drummond and Liu (2013), who assess the impact of China’s
investment growth on export growth using data for 174 countries covering the period 1995–2012.

www.economics-ejournal.org 7
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

where exportt stands for Tanzania’s exports; cdit for China’s domestic fixed asset
investment, which serves as a proxy for investment slowdown in China and the
accompanying contraction in demand for Tanzania’s exports; pricet for Tanzania’s
export commodity prices, included to capture the impacts on Tanzania’s exports of
falling global commodity prices; yworld
t
for world income, representing overall
external demand for the country’s exports; and ∆ denotes the first difference
operator. Eqs. (1) and (2) are estimated to quantify spillovers into Tanzania’s
exports from an investment deceleration in China, which are measured by the key
parameters: 𝜃1 and 𝜃1∗. As discussed in Section 4, the formulation of the
cointegrated VAR model allows us to estimate both the long-run (captured by 𝜃i in
Eq. (1)) and short-run (represented by 𝜃i∗ in Eq. (2)) effects of changes in the
respective explanatory variables. 𝜃1 and 𝜃1∗ are expected to be positive as higher
domestic investment in China tends to increase demand for Tanzania’s exports,
with an additional impact operating through higher export prices. We expect 𝜃2 and
𝜃2∗ to be positive because higher export prices would translate into higher value of
exports. The coefficients likely represent causal relationships as it is highly
implausible that the level of domestic investment in China and commodity prices
are affected by Tanzania’s level of exports.
The focus is on domestic investment because it supposedly better captures
China’s structural shift away from investment-driven growth and its potential
economic slowdown. As touched upon previously, an investment boom
underpinned China’s breakneck growth over the past decade and the ensuing rapid
growth in import demand for primary commodities, which are the mainstay of
Tanzania’s exports. In addition, China’s rebalancing towards a consumption-led
economy is likely to be reflected in a slowdown in investment spending (Lakatos et
al., 2016). IMF (2012) finds that economies within China’s supply chain and those
with less diversified commodity exports are the most vulnerable to deceleration in
China’s investment growth. Real output might serve as an alternative indicator of
economic activity in China; however, it exhibits limited time variation compared to
investment. Thus, investment may better help explain movements in exports, which
vary substantially over time. However, as shown later in the paper, using output
instead of investment leaves our main conclusion broadly unaffected. Note that we
use total exports in nominal terms to capture both the volume and price effects on
Tanzania’s exports of China’s domestic developments.7

7
As trade in goods tends to generate trade in services, such as in telecommunication and
transportation sectors, the focus is on exports of goods and services rather than only exports of goods.

www.economics-ejournal.org 8
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

3.2 Volatility in financial markets and growth

The contribution of capital flows to the Tanzanian economy and the impact of its
volatility on GDP growth is assessed by estimating the following model (hereafter
referred to as Model 2):

yt = 𝛿0 + 𝛿1 capt + 𝛿2 invt + 𝛿3 ext + 𝜀𝑡 (3)

∆yt = 𝛿0∗ + 𝛿1∗ ∆capt + 𝛿2∗ ∆invt + 𝛿3∗ ∆ext + 𝛿4 ∆volt (4)

where yt stands for Tanzania’s real GDP; capt for net private capital inflows to
Tanzania, included to measure the contribution of international capital flows to
national output and thus, by implication, the loss in real output associated with
lower capital flows stemming from tightened global financial conditions; invt and
ext for domestic investment and exports respectively, which constitute additional
important determinants of real income; and ∆vol𝑡 for the standard deviation of net
capital flows and captures volatilities in capital flows. 8
As noted above, capital flow is the main transmission mechanism through
which turbulences in global financial markets might ripple into the domestic
economy. Accordingly, perturbations to the economy arising from volatility in
capital flows are modeled in Eq. (3) via the spillovers of changes in capital inflows
into real national income and in Eq. (4) via the growth impact of capital flow
volatility. 𝛿1 and 𝛿1∗ are expected to be positive because an increase in capital
inflows is widely believed to be beneficial to recipients through promoting
productive investments, enhancing efficiency, and facilitating technology adoption.
However, the impact of capital flows may also depend on their size and volatility,
with flows being more beneficial to countries that have reached a certain threshold
of financial and institutional development. We expect 𝛿2 and 𝛿3 to be positive for
obvious reasons. 𝛿4 is expected to be negative as capital inflow surges and
disruptive outflows carry risks to economies, particularly to low income countries
like Tanzania.

3.3 Currency movements and inflation

The impact of changes in the nominal exchange rate on inflation is investigated

based on the following model (hereafter referred to as Model 3):
8
Note that volatility enters only Eq. (4) because it is of particular interest to assess the impact of
volatility in capital flows on economic growth. Volt does not enter Eq. (3) as real GDP changed
slowly over time while volatility in capital flows saw marked fluctuations.

www.economics-ejournal.org 9
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

∆pt = 𝜙0 + 𝜙1 neert + 𝜙2 ∆poil

t
+ 𝜙3 mt + 𝜙4 ∆pfood
t
+ 𝜀𝑡 (5)

where ∆p𝑡 stands for the inflation rate; neer𝑡 for nominal effective exchange rate
(henceforth NEER); 9 ∆poil
𝑡
for world oil price inflation, controlling for the impact
of supply shocks; m𝑡 for broad money supply or M2, accounting for the effect of
monetary policy shocks; and ∆pfood 𝑡
for global food price inflation. 𝜙1 is the
parameter of particular interest and captures the effect of movements in NEER on
inflation. A negative coefficient on 𝜙1 would be theoretically consistent as large
nominal depreciation (i.e. a decline in neer𝑡 ) triggers inflationary effects by, among
others, increasing import prices. 𝜙3 > 0 suggests that an excessive growth in
aggregate demand induced by higher money supply increases domestic prices and
fuel inflation, a phenomenon oft-described as ‘too much money chasing too few
goods’. The coefficients on oil and food price inflation are also expected to be
positive. Lower global oil and food prices have purportedly played a significant
role in containing inflation within single-digits territory and might have partly
offset currency depreciation-induced hikes in inflation. This is essentially an
empirical question and needs to be settled based on thorough empirical analysis.

4 Data
4.1 China economic slowdown and commodity prices

The data are annual observations for the period 1990−2014 and comprise the
variables: exports of goods and services (denoted with ex); China’s gross fixed
capital formation (cdi); export price index (price); and world GDP (yworld). They
are obtained from the World Bank’s World Development Indicators and the Bank
of Tanzania. All variables except export are at constant market prices, i.e. they are
adjusted for price (or inflation) effects. In addition, all data are given in US
Dollars. We opt for the nominal value of exports; however, the main conclusion of
this analysis is robust to using exports in constant prices instead. Looking at the
effect on exports in constant prices would be tantamount to disregarding the impact
of China’s economic slowdown operating via lower commodity prices. It bears
noting that a valid assessment of the impact of China’s slowdown is difficult and
somewhat premature because China’s strong economic ties with Tanzania are a
relatively recent phenomenon. Thus, we focus on the period since 1990.

9
Nominal effective exchange rate is computed as the weighted average of the bilateral exchange rates
between the country and each of its trading partners, where the weights are the respective trade shares
of each partner.

www.economics-ejournal.org 10
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Although important variables are omitted from our analysis, this does not, in
general, invalidate the long-run estimates. The reason is that cointegration property
is invariant to changes in the information set, i.e. a long-run relation detected
within a given set of variables will also be found in an enlarged variable set
(Johansen, 2000). Note that we also estimate an extended model that includes net
FDI flows to test the hypothesis that the level of FDI inflows to Tanzania is
influenced by export commodity prices. The advantage of gradually expanding the
information set is twofold. First, it greatly facilitates the identification of long-run
relations. Second, it enables an analysis of the sensitivity of the results associated
with the ceteris paribus assumption ingrained in the smaller model. The graphs of
the variables in levels and first differences are shown in Appendix Figures 1 and 2,
respectively.

4.2 Volatility in capital flows

The model linking capital flow volatility and economic growth is based on data
spanning the period 1980–2014 and includes the following variables: real GDP
(denoted with y𝑡 ); net private capital inflows (capt ); exports of goods and services
(ext ); and gross domestic investment (invt). The data were extracted from the
World Bank’s World Development Indicators and IMF’s Balance of Payments
Statistics. Because private capital inflows rose dramatically only over the past
couple of decades, we check the robustness of the key findings by restricting the
sample to cover only the period 1990–2014. However, focusing on the last two
decades would likely increase the statistical significance of these variables.
Appendix Figures 3 and 4 present the graphs of these variables both in levels and
first differences, respectively.

4.3 Currency depreciation and inflation

For this analysis, we use monthly data for the period from January 2013 to January
2015 and include the variables: inflation rate (denoted with ∆p𝑡 ) (change in the log
of consumer price index (CPI)); nominal effective exchange rate (neer𝑡 ); broad
money or M2 (m𝑡 ); world oil price inflation (∆poil t
) (change in the log of crude oil
food
price index); and global food price inflation (∆pt ). A broad monetary aggregate,
such as M2, is likely to have a closer link with the inflation rate than base money. 10

10
The base money includes currency in circulation and reserve balances held with the central bank by
depository institutions.

www.economics-ejournal.org 11
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Nonetheless, the central bank can only control this broader aggregate indirectly by
manipulating the monetary base. The analysis controls for the effects of monetary
policy using broad money; however, using rather the monetary base or narrow
money (M1− currency in circulation outside banks and demand deposits of
Tanzanian residents with banks) does not significantly matter for the conclusions
from this analysis.
Our analysis omits real GDP, a potential indicator for real economic activity
because data on this variable are not available on a monthly basis. However, the
key findings of the analysis would generally remain unchanged if we included real
GDP for at least two reasons. First, due to invariance of cointegration analysis to
expansions in the variable set, adding real GDP would leave the core (long-run)
results from the smaller model more or less intact. Second, since our sample covers
only the last three years and given overall economic activity was fairly stable
during this period, higher (or lower) inflation rate seems less likely due to stronger
(or weaker) economic growth and more likely due to increases (or declines) in
global energy and food prices, less (or more) prudent monetary policy, and faster
(or slower) depreciation of the Shilling (see World Bank, 2015b, 2016). The graphs
of these variables in levels and first differences are shown in Appendix Figures 5
and 6, respectively.

5 Model specification
5.1 The Cointegrated VAR model

Macroeconomic time-series data are typically characterized by path dependence,

interdependence, unit-root non-stationarity, structural breaks, as well as shifts in
equilibrium means and growth rates. To be a satisfactory benchmark a statistical
model needs to simultaneously address these data features. Path dependence would
point to a time-dependent process such as the autoregressive model, variable
interdependence to a system-of-equations approach, and unit-root nonstationarity to
cointegration. The cointegrated VAR model satisfactorily deals with these salient
features of the data. Unlike other approaches in which data are constrained in pre-
specified directions and are assigned an auxiliary role of ‘quantifying’ the
parameters of an ad hoc theoretical model, the cointegrated VAR methodology
uses strict statistical principles to extract out meaningful relations from the data
(Hoover et al., 2008; Spanos, 2009).
The analysis uses the cointegrated VAR model, which is structured around 𝑟
cointegration relations (the endogenous or pulling forces) and p − r stochastic

www.economics-ejournal.org 12
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

trends (the exogenous or pushing forces). The baseline model is specified with one
lag, a linear trend restricted to the cointegration (long-run) relations and a number
of dummy variables to be explained below: 11

∆xt = 𝛼β∗ x∗t−1 + Γ1 ∆xt−1 + ΦDs,t + 𝜙Di,t + 𝜇0 + 𝜀t

where x∗t−1 = (𝑥t−1 , t, t𝑦𝑦 ) is a p-dimensional vector of variables defined in Section

4; 12 𝛼 represents adjustment (or error-correction) coefficients (denoting the speed
of adjustment to equilibrium); β∗ = (β′ , β1 , β11 ) is a vector of coefficients to the
long-run relations; β′ xt are r long-run relations; t is a linear trend (1,2,3,…), β1 is a
r-dimensional vector of trend coefficients of the long-run relations; tyy is a broken
linear trend (…0,0,0,1,2,3,…) starting in the year yy and restricted to the long-run
relations (see discussion below), β11 measures the change in the linear trend
coefficient (or slope) of the long-run relations that ensued the extraordinary event
in yy; D𝑠,𝑡 is an unrestricted step dummy (0,0,0,1,1,1) starting in yy and controls for
shifts in growth rates as well as changes in the means of long-run relations; Di,𝑡
(…,0,0,0,1,0,0,0,…) is a permanent impulse dummy and accounts for an
unanticipated one-period shock effects in yy; Φ and 𝜙 are coefficients to the step
and impulse dummies, respectively; 𝜇0 is a vector of constant terms; 𝜀𝑡 is a 𝑝 × 1
vector of error terms; and ∆ is the first difference operator. Some of the variables in
our models experienced significant breaks in their long-run trends (and thus mean-
shift in their growth rates), which were modelled by allowing for a piecewise linear
trend, β1 t + β11 t𝑦𝑦 , in the long-run relations and a step dummy, Ds,t , in the
equations, ∆x𝑡 . Since all variables are in logs, their differences represent growth
rates.

11
The software package OxMetrics (Doornik and Hendry 2001) was used to carry out all
computations.
12
Using this model generally requires that the vector process 𝑥𝑡 is at most 𝐼(1). (The series 𝑥𝑡 is said
to be integrated of order d, 𝑥𝑡 ~𝐼(𝑑), if it is stationary when differenced 𝑑 times, which means 𝑥𝑡
contains 𝑑 unit roots.) The order of integration of 𝑥𝑡 is initially determined by applying the univariate
augmented Dickey-Fuller (ADF) test for the variables in our models. Given most of the variables are
trending the relevant alternative hypothesis in most cases is trend stationarity. In addition, for some of
the variables, the univariate unit-root tests were applied also allowing for a change in trend slope at
dates identified using the statistical procedure in Doornik et al. (2013). Most of the variables were
found to be unit-root non-stationary, although some were stationary when the test included a trend
term (and/or a change in trend slope). Next, we determined the order of integration and cointegration
based on multivariate VAR-based unit-root test. Neither the choice of full rank (data in levels are
stationary) nor zero rank (data are nonstationary but not cointegrated) was supported by the statistical
tests in all cases, suggesting that 𝑥𝑡 ~𝐼(1). The unit-root test results are not reported here, but can be
made readily available upon request from the author.

www.economics-ejournal.org 13
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

5.2 Specification tests

The VAR model is derived under the assumption of constant parameters and
multivariate normality. Although parameter stability can be assessed using
recursive test procedures, the small number of observations at our disposal
circumscribes the power of available recursive procedures. However, since both
parameter non-constancy and non-normal errors are often associated with periods
of political and economic turbulence, such as supply shocks, war, severe droughts,
civil unrest, and policy interventions, we improve parameter stability and mitigate
on-normality by controlling for the most dramatic events using several dummy
variables. In fact, some of the variables feature few extraordinarily large
observations incongruous with the normality assumption.
First step in the empirical analysis is to determine the lag length of the VAR
model. Statistical tests indicate that there is no evidence of residual autocorrelation
in the VAR(1) (i.e. a model with a leg length of 1) for all models. Accordingly, the
lag length was truncated to 1. However, the results obtained allowing for two lags
are by and large similar with the ones presented below. Provided that there are no
signs of autocorrelation in the residuals, and given the relatively large number of
variables and small size of our sample, the VAR(1) model is a satisfactory and
parsimonious representation of the variation in the data.
In Model 1, the following dates were classified as outlying observations: 1989,
1998, and 2009. 13 These outliers correspond to observations with standardized
residuals larger than 3.0, i.e. |𝜀 ⁄𝜎𝜀 | ≥ 3.0, which is the standard criteria for
identifying an outlier. 14 An algorithm searching for breaks and aberrant
observations developed in Doornik et al. (2013) was used to determine the
existence, timing, and significance of outliers, and shifts in mean growth rates. The
year 1989 corresponds to the sharp economic downturn in China due to civil unrest
and the subsequent economic sanctions several countries imposed against it. 15 1998
coincides with the decline in world commodity demand as a result of the Asian

13
Global oil and food price inflation, and world income were modelled as weakly (long-run)
exogenous variables for two reasons. First, given the small size of the sample, we preserve degrees of
freedom by treating these variables as exogenous. Second, it is highly implausible that the long-run
paths of the variables are affected by any of the variables in our empirical models. However, the main
results of this paper prove robust to relaxing this assumption.
14
Contrary to the case in static regressions, the dummies do not eliminate the corresponding
observations. The dummies account for unanticipated shocks and given that these are no longer
unanticipated in the next period, their lagged effects on the system are accounted for by the dynamics
of the model.
15
In 1989 and 1990, China’s real GDP growth stood at about 4 percent, which represent the lowest
growth rates over the past several decades. In addition, gross investment fell by about 15 percent in
1989.

www.economics-ejournal.org 14
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

financial crisis of 1997−1998, which led to a significant drop in Tanzania’s export

prices. The year 2009 marks the global economic slump, also dubbed the Great
Recession, which took a heavy toll on most advanced economies and saw world
GDP drop by around 2 percent.
We also spotted a change in trend slope in (and thus a shift in the mean growth
rate of) export price in 2002. As discussed in Section 2, global commodity prices
moved onto higher growth trajectory in 2002, which lasted more than a decade and
has often been referred to as commodities “super-cycle”. During this super-cycle
period, Tanzania’s export prices experienced hefty growth. We control for this
event using a broken linear trend in the long-run relations and a step dummy in the
equations in 2002. The location shift in growth rates is shown in Appendix Figure
7. In sum, the specification for Model 1 includes a linear trend and a broken linear
trend (with a change in trend slope in 2002) restricted to the long-run relations, an
unrestricted shift dummy in 2002 (which controls for the shift in growth rates as
well as the change in means of long-run relations), and an unrestricted impulse
dummies (accounting for an unanticipated one-period shock effects) in 1998 and
2009. 16 In addition, the baseline model treats world GDP as a weakly exogenous
variable. (Appendix Table 3 shows that the key results are robust to relaxing this
assumption.) Further, the lag length is set equal to k = 1 in levels. 17
In Model 2, the diagnostic tests detected a structural break in real GDP in 2001
as well as a number of outlying observations. The former captures the relatively
higher and sustained economic growth Tanzania enjoyed since the early 2000s
(Appendix Figure 8). Average annual GDP growth exceeded 6 percent since 2001,
which constitutes a remarkable break from the past, with growth averaging less
than 3 percent during 1980–2000. Tanzania experienced a dramatic increase in
investment in the second half of the 1980s following the adoption of the Economic
Recovery Program, primarily fueled by surges in foreign aid inflows. The
investment spikes in 1987 and 1990 were controlled for using impulse dummies. In
addition, the observation 1985 was classified as ‘too large’, which is associated

16
The analysis controls for the most dramatic events using different types of dummy variables. For
example, a shift in the equilibrium mean can be captured by a step dummy, 𝐷𝑠 𝑦𝑦𝑡 , defined as
(0,…0,0,0,1,1,1,…,1), while a one-period shock effect can be accounted for by an impulse dummy,
𝐷𝑖 𝑦𝑦𝑡 , defined as (0,…,0,0,0,1,0,0,0,…,0). In addition, in models with changes in trend slopes in the
long-run relations, there is a need to additionally account for the change in underlying trends (and
thus the corresponding shift in long-run growth rates). Such events were modelled using a broken
linear trends (𝑡𝑦𝑦) in the long-run relations, 𝛽′ 𝑥𝑡 , and a step dummy (𝐷𝑠 𝑦𝑦𝑡 ) in the equations, 𝑥𝑡.
17
However, the results obtained for k = 2 are by and large similar with the ones presented below.
Provided that there are no signs of autocorrelation in the residuals and given the large-dimensional
systems and the small sample that we have at our disposal, the VAR(1) is a more parsimonious
representation of the data generating process (DGP).

www.economics-ejournal.org 15
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

with the sharp (relative) increase in net private capital inflows. In a nutshell, Model
2 was specified to allow for a linear trend and a broken linear trend (with a change
in trend slope in 2001) restricted to the long-run relations, an unrestricted shift
dummy in 2001 (which accounts for the mean shift in ∆𝑥𝑡 and controls for the shift
in means of long-run relations), and an unrestricted impulse dummies in 1985,
1987, and 1990. In addition, statistical tests indicated that one lag was the optimal
lag length and k was truncated to 1 accordingly.
Model 3 became well-specified when we allowed for a broken linear trend in
2013(7) (i.e. the seventh month of 2013) and 2014(8), and the impulse dummies:
Di 12.6t (where 12.6 denotes the sixth month of the year 2012), Di 13.1t, and
Di 13.4t. The trend break in 2013(7) represents the shift in the growth path of
inflation, which assumed astronomical proportions in 2012 and for most of 2013,
whereas it receded to reasonable single-digit rates over the past two years and half.
The broken trend in 2014(8) accounts for the change in the long-run trend
underlying money supply. Money supply increased steeply until late 2014, after
which it shifted to a noticeably lower growth path. See Appendix Figures 9 and 10.
Further, there is evidence of considerable seasonality in the monthly data, which
we accounted for using seasonal dummies. To sum up, the specification for Model
3 includes: a linear trend and broken linear trend (with changes in trend slope in
2013(7) and 2014(8)) restricted to the cointegration space, an unrestricted shift
dummy in these periods, and an unrestricted impulse dummies in 2012(6), 2013(1),
and 2013(4). In addition, the lag length for Model 3 was set equal to 2. Global food
and oil price inflation rates were modeled as weakly exogenous variables in the
baseline model.
We now shed some light on how the above-mentioned break points were
identified and accounted for. As alluded to above, we identified the break dates
based on a priori knowledge on the timing of special events, a graphical inspection
of the data, as well as a statistical test for the presence of trend and level shifts in
the data. The broken trend possesses the most significant coefficient at those
periods and accordingly the models were specified with a change in trend slope at
these points. The hypotheses that Tanzania has had no statistically significant shift
in the mean growth rates of the series at the specified dates were strongly rejected
(p-value: 0.00).
Turning to diagnostic tests, I identified and tested for trend breaks using
univariate as well as multivariate statistical procedures. In particular, an algorithm
searching for breaks developed by Doornik et al. (2013) and the procedure in
Hungnes (2005) were used to determine the existence, timing, as well as the
significance of breaks in mean growth rates. Sustained shifts in growth rates were
defined following Hausmann et al. (2005): (i) For a shift in mean growth rate to be

www.economics-ejournal.org 16
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

categorized as a growth turnaround it should be sustained for at least 8 years and

the change in growth rate has to be at least 2 percentage points; (ii) A variables can
experience more than one instance of growth turnaround as long as the dates are
more than 5 years apart; (iii) Trend breaks were selected at 1% ‘target size’ 18 (i.e.
𝛼 = 0.01) in the Autometrics options in OxMetrics 7 (see Doornik et al., 2013).
Note that we perform a sensitivity analysis to examine if the estimates based on the
statistically and economically most credible break date are fairly robust to
alternative candidate break points in the vicinity of the first-best break point. We
find that the main conclusions of this paper prove robust to changes in the break
dates.
In modelling structural breaks, the paper draws on the conventional
(multivariate) cointegration approach in Johansen et al. (2000) and Hungnes
(2010), which accommodates different types of structural breaks. Specifically,
using such a multivariate framework, hypothesis testing on breaks in trend slopes
(or shifts in growth rates) can be formulated and properly tested. A potential
drawback of a system-of-equations approach is that the trend breaks are assumed to
occur at the same date for all series. An alternative would be to use a univariate
approach and apply some variant of the method proposed by Perron (1989).
However, in our case, the use of a single equation model would be more restrictive
and hard to justify in the face of overwhelming evidence for the existence of more
than one cointegration relations in all three models. In addition, Bai et al. (1998)
show that there are substantial gains in precision from using multivariate models in
which several variables are modelled as cointegrated system. The use of multiple
series sharpens inference about the existence and dates of shifts in the mean levels
(Bai et al., 1998, pp. 420). In other words, a break in mean growth rates might be
more readily detected and estimated in a multivariate setting including variables
that are purportedly co-moving. In some respects, our approach is similar to that of
Hausmann et al. (2005), Wacziarg and Welch (2008), and Jones and Olken (2008),
who identify episodes of sustained shifts in growth rates and examine explanations
for such transitions.
All empirical models inherently approximations of the actual data generating
process and we now turn to assessing if the models described in the previous
section are reasonable approximations. Table 1 reports multivariate specification
test results as well as univariate statistic corresponding to normality tests for all
three models under consideration. With the deterministic specifications and the
dummies included, the models discussed above pass most of the specification tests

18
The target size determines the significance level below which a break is not kept in the model.

www.economics-ejournal.org 17
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

and describe the data reasonably well. No serious deviations from the assumptions
of residual independence and normality was detected.
In the three models, the null of normal errors was only borderline accepted.
However, a look at the univariate test statistic indicates that normality was
accepted in all equations, albeit with a relatively small p-value for some of the
variables due to excess kurtosis (long tails). This, coupled with the absence of
autocorrelation, seems to suggest that the result of the multivariate test is a finite
sample phenomenon given that we have a small sample and a large number of
variables. 19 A look at the univariate tests statistic in Table 1 clearly indicates that
normality was borderline accepted due to excess kurtosis whereas all individual
equations have a skewness close to zero. We have gone to great lengths to ensure a
model set up where multivariate normality is accepted with a higher p-value by,
inter alia, estimating a partial model conditioning on weakly exogenous variables
and changing the sample period. All these avenues, however, lead to similar
conclusions. The multivariate tests of no autocorrelation were not rejected in all
except Model 1, although with relatively small p-values. Note that the main

Table 1. Model specification tests

Model 1 Model 2 Model 3
Var. p-value Var. p-value Var. p-value
Normality* 0.01 0.07 0.05
Ex 0.58 y 0.57 ∆p 0.39
CDI 0.07 Cap 0.16 neer 0.72
Price 0.54 Inv 0.14 m 0.12
Ex 0.08
Skewness Ex 0.09 y 0.40 ∆p 0.61
CDI 1.36 Cap 0.56 neer 0.00
Price 0.11 Inv 0.13 M 0.09
Ex 0.33
Excess kurtosis Ex 3.21 y 2.88 ∆p 3.56
CDI 5.02 Cap 4.56 neer 2.20
Price 3.81 Inv 3.53 m 4.63
Ex 4.29
Autocorrelation 0.09 0.11 0.17
ARCH 0.21 0.46 0.59
Note: These figures represent p-values. The p-values measure the degree to which the null
hypothesis is accepted: The higher the p-value, the more strongly the null hypotheses of
normal errors, no autocorrelation, and no ARCH effects are accepted. *The p-values in
bold face correspond to tests of multivariate normality while those under the multivariate
test results represent univariate tests statistic for each of the three models.

19
The main conclusions of our analysis are sufficiently robust to steps that might circumvent the
problem, such as increasing the lag length.

www.economics-ejournal.org 18
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

conclusions from our analysis prove robust to steps that might circumvent the
problem, such as increasing the lag length. In addition, although there are some
signs of moderate ARCH effects and excess kurtosis (long tails), cointegrated VAR
results are reasonably robust to such effects (Gonzalo, 1994; Rahbek et al., 2002).
Having established an adequate statistical description of the data, the next step
is determining the cointegration rank. The cointegration rank classifies the data into
𝑟 long-run relations towards which the process is adjusting (the pulling forces)
and 𝑝 − 𝑟 relations which are pushing the process (the exogenous forces). The
choice of rank is made based on a range of statistical criteria, such as the trace test,
the largest unrestricted root of the characteristic polynomial for a given 𝑟, the t-
ratios of the α coefficients for the 𝑟 𝑡ℎ cointegration vector, and the graphs of the
𝑟 𝑡ℎ cointegration relation. Table 2 reports the p-values of the trace test (𝜏), the
largest unrestricted characteristic root (𝜌) and the largest 𝑡-value of the 𝛼
coefficients (α )̂ . The test results indicate that 𝑟 ∗ is the statistically mostcredible
(first-best) choice of rank for all three models. 20 This suggests that there exist two
long-run relations among the variables in our models. The choice of rank is
conventionally made based on the trace test.
However, because the trace test suffers from substantial power problem when
the size of the sample is small, we also base the choice of rank on the significance
of the 𝛼 coefficients, the characteristic roots of the model, and the graphs of the
long-run relations (Juselius, 2006: Chapter 8.5). For the choice of r = 3, the largest
unrestricted root for the three models seems a bit far from the unit circle whereas
the t-values of 𝛼 reveal that there is no significant adjustment to the last two
cointegration vectors. In other words, the strong persistence and much less
significant adjustment coefficients for the third cointegrating vector might be used
as a safeguard against including it in the stationary part of the model. In addition,
the graphs of the recursively calculated trace tests exhibit pronounced linear
growth in the first two cointegration relations, but much less so in the last two,
although the picture is not as clear cut for Model 3. Similarly, a glance at the
graphs of the cointegration relations reveal that the last two cointegration vectors in
Models 1 and 2, and the last three in Model 3 show distinct non-stationary
behavior, pointing toward r = 2. Hence, taking all these into consideration, we
consider the choice of r = 2 to be a reasonable choice. It is important to note,
however, that the main conclusions of this paper are fairly robust to altering the
cointegration rank. See discussion in Section 6.

20
These results are not reported here, by can be made readily available upon request form the author.

www.economics-ejournal.org 19
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Table 2. Determination of cointegration rank

Trace test (𝜏), characteristic roots (𝜌�), and t-values of 𝛼 (𝛼�)
𝜏 𝜌� 𝛼�
r∗ − 1 r∗ r∗ + 1 r∗ − 1 r∗ r∗ + 1 r∗ − 1 r∗ r∗ + 1
∗ 0.07 0.57 0.83 0.67 3.2 1.7
Model 1 (r = 2) 0.39 0.67 3.4
Model 2 (r∗ = 2) 0.00 0.08 0.15 0.71 0.71 0.55 7.7 6.2 2.5
Model 3 (r∗ = 2) 0.00 0.01 0.04 0.83 0.72 0.50 4.4 5.7 2.1

Note: The figures represent p-values of the trace test (𝜏), the largest unrestricted characteristic root (𝜌�), and the
largest t-value of the error-correction coefficients (𝛼�).

6 Results
This section discusses the identified structures of long-run equilibrium
relationships for Models 1 – 3. When interpreting the results in this section it
should be borne in mind that a cointegration relation only measures the association
between the variables over the long-run and as such does not say anything about
causality. To say something about causality, we need to combine the cointegration
coefficients, β, with the adjustment coefficients, 𝛼. For example, the hypothetical
cointegration relation (x1,𝑡 − 𝛽1 x2,𝑡 )~𝐼(0) describes a positive comovement
between x1,𝑡 and x2,𝑡 . If the adjustment coefficient 𝛼1 , of x1,𝑡 , is negative and
significant but the adjustment coefficient corresponding to x2,𝑡 is insignificant, i.e.
𝛼2 = 0, we can say that the direction of causality runs form 𝑥2,𝑡 to 𝑥1,𝑡 , i.e. 𝑥1,𝑡 =
𝛽1 𝑥2,𝑡 + 𝜇𝑡 .
However, the interpretation becomes less straightforward in terms of sign
effects as the number of variables in a long-run relation increases. As alluded to
above, when discussing the empirical results below, it should be noted upfront that
the small number of observations at our disposal circumscribes the power of some
multivariate test statistic, such as the trace test and recursive tests of parameter
stability. Thus, the results below need to be interpreted bearing in mind these
caveats. In particular, considering the volatile history of Tanzania, some of the
estimated coefficients may represent average historical effects.

6.1 China’s economic slowdown and commodity prices

We initially discuss the baseline ordinary least squares (OLS) estimates. However,
some of the independent variables may be correlated with a number of other
variables, thereby not warranting causal interpretation of the estimates. In addition,
the variables in our model are quite persistent over time. Thus, OLS might produce

www.economics-ejournal.org 20
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

unreliable results, which prompts the need for a statistical model that addresses this
data feature. Further, omitted variables and multicollinearity problems could render
the baseline estimates biased. Therefore, we also estimate the models using the
cointegration VAR methodology. Unlike the OLS regression, collinearity between
the variables does not result in imprecise estimates of the long-run relations based
on the cointegrated VAR model. The reason for this is that, unlike the case with a
regression analysis in levels, the cointegrated VAR formulation more or less
circumvents the multicollinearity problem by transforming trending variables into
stationary differences, ∆x𝑡 , and stationary long-run relations, β′𝑥𝑡 (Juselius, 2006).
The baseline OLS results are reported in Columns 1 (long-run) and 4 (short-
run) of Table 3. cdi𝑡 possesses a positive coefficient estimate, suggesting that an
increase in China’s domestic investment is associated with higher Tanzanian
exports. A 1 percentage point (ppts) increase in China’s investment growth is
correlated with 0.89 ppts increase in export growth. We now resort to the estimates
from the cointegration analysis. Table 3 reports the identified structure of two long-
run relations, which was accepted based on a high p-value of 0.74. The estimated
structure is generically, empirically, and economically identified as defined in
Johansen and Juselius (1994).
The first long-run relation is between exports value, China’s domestic
investment, and prices. The estimated error-correction coefficients reveal that

Table 3. Impact of China’s slowdown and falling commodity prices (1990−2014)

Long-run analysis Short-run analysis
(1) (2) (3) (4)
OLS Cointegrated VAR Cointegrated OLS
(Accepted with a p-value of 0.74) VAR
Independent 𝑒𝑒𝑡 Long-run relations Error correction Dep. variable ∆𝑒𝑒𝑡 ∆𝑒𝑒𝑡
Var. → Coefficients →
Dep. var. ↓ Var. 𝛽̂1 𝛽̂2 𝛼�1 𝛼�2
𝑐𝑐𝑐𝑡 0.87 𝑒𝑒𝑡 1.00 − −0.78 −0.75 ∆𝑒𝑒𝑡−1 0.27 0.28
(3.14) (−4.70) (−4.54) (1.95) (1.52)
𝑝𝑝𝑝𝑝𝑝𝑡 0.42 𝑐𝑐𝑐𝑡 −0.57 −0.97 ∗ ∗ ∆𝑐𝑐𝑐𝑡−1 0.60 0.89
(3.98) (−6.35) (−3.05) (3.46) (2.98)
𝑦𝑡𝑤𝑤𝑤𝑤𝑤 3.32 𝑝𝑝𝑝𝑝𝑝𝑡 −0.65 1.00 ∗ −0.45 ∆𝑝𝑝𝑝𝑝𝑝𝑡 − 0.14
(2.42) (−3.51) (−1.90) (1.04)
𝑡 −0.21 𝑦𝑡𝑤𝑤𝑤𝑤𝑤 − −4.23 ∗ ∗ ∆𝑦 𝑤𝑤𝑤𝑤𝑤 1.49 1.85
(−2.72) (−3.85) (1.64) (1.22)
𝑡2002 0.05 𝑡 − 0.30
(2.12) (3.65)
𝑡2002 − −0.11
(−5.15)

Note: t-values in parentheses. *Denote insignificant adjustment coefficients (𝑡-value less than 1.80).

www.economics-ejournal.org 21
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

short-run adjustment occurs only through changes in exports, signifying its

importance as an export long-run relationship:

ex𝑡 = 0.57 cdi𝑡 + 0.65 price𝑡

The results suggest that ceteris paribus China’s domestic investment and export
prices make positive contribution to long-run movements in exports. Specifically,
the estimates, which represent causal effects, suggest that a 1 percent contraction in
domestic investment in China would lead to a drop in Tanzania’s exports of about
0.57 percent. This is consistent with the fact that an investment boom buoyed up
China’s impressive growth and that this was followed by burgeoning import
demand for primary commodities, which account for the lion’s share of Tanzania’s
export revenues. Conversely, the estimates reflect that a slower, more balanced,
growth in China has depressed global demand for commodities and hence held
back lower Tanzanian exports. The short-run results (Column 3 of Table 3)
indicate that a 1 ppts decline in China’s investment growth is associated with 0.60
ppts decrease in Tanzania’s export growth.
The second long-run relationship describes a strong association between export
prices, China’s domestic investment, and world income. The adjustment coeffi-
cients show that only export price is error-correcting to this equilibrium
relationship:

price𝑡 = 0.97 cdi𝑡 + 4.23 yworld

𝑡
− 0.30 t + 0.11 t2002

We find that increases in China’s domestic investment and world income are
associated with higher prices for Tanzania’s export commodities. The impact of a 1
percent investment slowdown in China is to reduce Tanzania’s export prices by
nearly 1 percent. This is to be expected because Tanzania is one of the countries
within China’s supply chain and a net exporter of commodities, the prices of which
have been driven by China’s domestic economic developments for more than a
decade. 21 A case in point is the recent drop in China’s gold imports from Tanzania
and the steady decline in the prices of gold over the last three years.22 In addition,
we estimate that an additional 1 percent increase in world income is associated
with an increase in export prices of about 4 percent. Vulnerability to wild
fluctuations in world commodity prices remains to be the Achilles' heel of the
21
As noted previously, the commodities super-cycle in the 2000s and early 2010s, and the coming to
an abrupt end of the seemingly-unstoppable surge in global commodity prices were partly triggered
by swings in China’s business cycle.
22
Slower metal-intensive investment in China was key in driving down base metal prices, which is
expected to continue throughout the growth transition (IMF, 2015).

www.economics-ejournal.org 22
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Tanzanian economy. Needless to say, channeling efforts toward diversifying the

export portfolio and markets destination, and improving the quality of existing
products can help the country mitigate headwinds from price swings and thus boost
its competitive standing. Altogether, the findings reflect the fact that Tanzania’s
exports are mainly composed of less diversified commodities, the prices of which
are generally determined in the global market and fall beyond the domains of
Tanzanian policy makers.
These findings are sufficiently robust to a battery of sensitivity checks.
Appendix Table 1 adds net FDI inflows to the data vector in Model 1 to examine
how direct investment flows to Tanzania are affected by fluctuations in commodity
prices. The results indicate that the first two long-run relations describe similar
export and price relationships as in Model 1, consistent with the invariance of
cointegration relations to expansions of the information set. The third long-run
relationship indicates that commodity prices are among the key determinants of
FDI inflows. Specifically, a 1 percent drop in export prices is associated with about
3 percent lower FDI inflows. This is to be expected as FDI flows to Tanzania have
increasingly focused on export-oriented production mainly related to investments
in extractive industries. Given that mineral and metal exports account for a good
portion of Tanzania’s exports, a significant decline in their prices might lead to a
scaling down of existing and new operations in the medium- to long-term.
As mentioned in Section 5, our baseline model specifies world income as a
weakly (long-run) exogenous variable. Allowing world income to enter Model 1 as
an endogenous variable, the analysis reaches the same conclusion as before
(Appendix Table 3). In addition, Appendix Table 2 shows that the results based on
the first-best choice of rank are robust to altering the cointegration rank to the
second-best alternative of r = 3. Further, as the sample size is small, the use of
dummies might have resulted in a considerable loss of degrees of freedom. Thus,
we redo the analysis excluding all dummy variables (Appendix Table 2). Our key
results remain broadly unchanged. Appendix Table 3 shows that the key
conclusions also hold up well to using GDP instead of investment as an indicator
for economic activity in China.
Figures 5 and 6 plot the generalized impulse response functions for exports in
response to a one standard error (se) shock to China’s domestic investment and
export commodity prices, respectively. Impulse response analysis describes the
knock-on effects on the system variables of a one se shock to a variable of interest,
assuming that the system is not hit by other shocks thereafter. It should be noted
that, unlike its orthogonalized counterpart, generalized impulse response is
invariant to ordering of the variables in the model. Nonetheless, for the effect of a

www.economics-ejournal.org 23
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Figure 5. Impulse resonse of export Figure 6. Impulse resonse of export to a

to a one SE shock to CDI one SE shock to export price
0,07
0,045
0,06 0,040
0,05 0,035
0,030
0,04
0,025
0,03 0,020
0,015
0,02
0,010
0,01 0,005
0 0,000
1 3 5 7 9 11 13 15 17 19 21 23 25 1 3 5 7 9 11 13 15 17 19 21 23 25

Horizon Horizon

shock to be economically meaningful, the shock should purely represent an

innovation to a particular variable instead of capturing a combined outcome of
correlated errors. This would, however, require the causal structure of the model to
be identified, rendering impulse responses susceptible to misinterpretation. Hence,
we complement our impulse response analysis with statistical and economic
evidence. In other words, we study the dynamic behavior of export following a one
se shock to the above-mentioned variables. We find that the contemporaneous
impact of a one se positive shock to domestic investment in China is an increase in
Tanzania’s exports of about 0.05 percent. The large current impact is accompanied
by a relatively modest increase, after which export gradually converges to its
higher equilibrium level of about 1.2. Turning to the impact of shocks to export
prices, the results show that export responds very slowly and the magnitude of the
effect is quite small in the first few periods. This seems to suggest that export
supply response to an increase in prices takes time to materialize. Following a
shock to prices, export starts to increase steeply after few periods and thereafter
converges smoothly to the long-run impact of 0.71 percent.

6.2 Volatility in capital flows

The OLS estimates in Column 4 of Table 4 show that higher volatility in capital
flows has a statistically insignificant impact on economic growth. The baseline
estimates are, however, valid only under fairly restrictive assumptions, which may
not be borne out by the data. Thus, we turn to the results based on the cointegrated
VAR model. For Model 2, the identified structure of two long-run relations was
accepted based a quite high p-value of 0.97. Table 4 reports the results. Note that,

www.economics-ejournal.org 24
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Table 4. Impact of volatility in capital flows (1980−2014)

Long-run analysis Short-run analysis
(1) (2) (3) (4)
OLS Cointegrated VAR Cointegrated OLS
(Accepted with a p-value of 0.97) VAR
Indep. 𝑦𝑡 Long-run Error correction Dep. Growth Growth
Var. → Relations coefficients Var. → ∆𝑦𝑡 ∆𝑦𝑡
Dep. var. ↓ Var. ̂
𝛽1 ̂
𝛽2 𝛼�1 𝛼�2
𝑐𝑐𝑐𝑡 0.01 𝑦𝑡 1.00 − −0.24 0.03 ∆𝑦𝑡−1 −0.63 −
(1.67) (−8.79) (4.70) (−4.13)
𝑖𝑖𝑖𝑡 0.21 𝑐𝑐𝑐𝑡 −0.04 −0.12 ∗ 2.74 ∆𝑐𝑐𝑐𝑡 0.001 0.004
(3.94) (−6.50) (−3.57) (−4.70) (0.30) (0.97)
𝑒𝑒𝑡 0.03 𝑖𝑖𝑖𝑡 −0.26 − 1.22 ∗ ∆𝑖𝑖𝑖𝑡 0.03 0.07
(1.11) (−14.95) (7.24) (2.18) (2.95)
𝑡 0.02 𝑒𝑒𝑡 − 1.00 0.74 −0.32 ∆Volt −0.01 0.002
(11.3) (−2.42) (−4.70) (−1.73) (0.36)
𝑡01 0.03 𝑡 − −0.07
(20.6) (11.09)
𝑡01 −0.03 −
(−11.87)

Note: t-values in parentheses. *Denote insignificant coefficients (𝑡-value less than 1.80).

although two-long relations were detected, we focus on the first long-run

equilibrium relationship, namely the GDP equation, because disentangling the
contribution of private capital flows to the national economy and the growth impact
of their volatility are of particular interest.
The first long-run relation comprises real GDP, net private capital inflows, and
investment:
y𝑡 = 0.04 cap𝑡 + 0.26 inv𝑡 + 0.03 t01

The estimated coefficients conform to a priori expectations. The results suggest

that capital inflows have marginally significant positive contribution to national
income. A 1 percent increase in net private capital inflows to Tanzania leads to a
0.04 percent increase in real GDP, which appears modest, albeit not negligible. The
positive impact of capital flows is in line with economic theory as capital inflows
are widely believed to benefit host countries through, inter alia, fostering
productive investments, unleashing efficiency, and accelerating the transfer of
technology. The small impact may be partly due to the fact that FDI flows, which
account for nearly all of the total private capital inflows to the country, grew
considerably only recently and that it may take a while before they translate into
higher level of output.
The short-run (cointegrated VAR) estimates in Column 3 of Table 4 show that
an increase capital flow volatility has a significantly negative, albeit modest,

www.economics-ejournal.org 25
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

impact on economic growth in Tanzania. In particular, a 1 ppts increase in the

volatility of net private capital inflows reduces growth by 0.01 ppts. The quite
modest effect seems to reflect the very small proportion of portfolio capital flows,
which tend to be more volatile and susceptible to changes in global financial
markets compared with FDI. The conventional wisdom suggests that, despite
theoretically sound arguments in favor of private capital flows, portfolio equity and
debt flows may pose substantial countervailing risks for developing economies as
they are often motivated by speculative considerations and thus prone to quick
reversals (Reinert et al., 2010). In Hausmann and Fernández-Arias (2000), short-
term capital flows are referred to as “bad cholesterol”. In contrast, FDI is chiefly
driven by long-term prospects and relatively irreversible in the short-run; hence
considered “good cholesterol”. Many developing countries, including Tanzania,
consider FDI as the private capital inflow of choice due to its purported resilience
in times of financial turbulence. This is in line with the growing consensus that low
income countries may need to reach a certain level of financial and institutional
development before they can start reaping the potential benefits of ‘relatively’
unfettered capital flows.

6.3 Currency depreciation and inflation

The baseline OLS estimates are reported in Column 1 of Table 5. We find that a 1
percent nominal exchange rate depreciation is associated with a higher inflation
rate of around 0.3 ppts. Moreover, money supply is negatively linked with
inflation, which appears counterintuitive at first glance. However, the estimated
coefficient does not represent causal relationship and likely captures the tendency
of the central bank to reduce money supply when inflation soars.
The identified long-run structure for Model 3 is shown in Column 2 of Table 5,
which was accepted based on a very high p-value of 0.93. Only the inflation
equation is discussed below as the overriding objective of the analysis is to
examine the impact of currency depreciation on inflation. The long-run relation for
the inflation rate is given by:

∆p𝑡 = −0.58 neer𝑡 + 0.21 ∆poil

𝑡
+ 0.51 ∆pfood
𝑡
+ 0.02t13(7)

The adjustment coefficients show that only inflation equilibrium-corrects

towards this cointegration relationship, suggesting that the direction of causality
goes from the NEER to inflation. Nominal depreciation is associated with higher

www.economics-ejournal.org 26
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Table 5. Nominal currency movements and inflation (Jan. 2013–Jan. 2015)

OLS Cointegrated VAR

(Accepted with a p-value of 0.93)
(1) (2)
Independent ∆𝑝𝑡 Long-run relations Error correction
variable → coefficients
Dep. var. ↓ Var. ̂
𝛽1 ̂
𝛽2 𝛼� 1 𝛼�2
𝑛𝑒𝑒𝑒𝑡 –0.32 ∆𝑝𝑡 1.00 –0.24 ∗
(–2.31) −
(–5.11)
𝑚𝑡 –0.16 𝑛𝑛𝑛𝑛𝑡 0.58 − 0.16 0.06
(3.93) (2.50) (3.66) (2.03)
∆𝑝𝑡oil 0.004 𝑚𝑡 − 1.00 0.52 –0.32
(0.211) (2.41) (–4.43)
∆𝑝𝑡food 0.096 ∆𝑝𝑡𝑜𝑜𝑜 –0.21 − ∗ ∗
(1.12) (–5.42)
𝑡 𝑓𝑓𝑓𝑓 −
–0.002 ∆𝑝𝑡 –0.51
(–4.46) (–2.51)
𝑡13(7) -0.012 𝑡 − –0.013
(–1.75) (7.95)
𝑡14(8) –0.014 𝑡13(7) –0.001 −
(–2.49) (2.30)
𝑡14(8) 0.03
− (10.30)

Note: t-values in parentheses. *Denote statistically insignificant adjustment coefficients

(𝑡-value less than 1.80).

inflation in Tanzania. Specifically, a 1 percent depreciation of the NEER leads to

0.58 ppts higher inflation. In addition, global oil and food price inflation rates
constitute important drivers of domestic inflation. A 1 ppts fall in global oil and
food price inflation rates results in lower inflation of about 0.21 ppts and 0.51 ppts,
respectively. The analysis finds no statistically significant link between changes in
money supply and inflation. Relaxing the assumption that oil and food price
inflation were exogenously determined does not significantly affect the baseline
findings.
All in all, the results suggest that the marked depreciation of the Shilling was
inflationary, although it was offset by the sharp slide in global commodity prices.
In other words, the somewhat low and stable inflation rates observed in recent
years were the result of a confluence of counteracting factors, including falling oil
and food prices, on the one hand, and the pronounced depreciation of the Shilling,
on the other. This implies that a possible concurrence of a resurrection in oil and
food prices, and significant currency depreciation might precipitate an inflationary
spiral. Short-run results (not reported here) suggest that nominal currency

www.economics-ejournal.org 27
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

depreciation has no significant contemporaneous and one-period lagged impact on

inflation, a result corroborated by findings from impulse response analysis. This is
not too surprising as the analysis focuses on a sample covering only three years and
relatively high-frequency monthly data, which might suggest that it takes more
than one month for a sizable nominal depreciation of the Shilling to take effect.

7 Concluding remarks
This paper conducted a thorough empirical analysis of the macroeconomic impacts
of recent global economic shocks in Tanzania. In particular, we set out to address
the intertwined questions of whether, and to what extent, China’s economic
slowdown, falling global commodity prices, and volatility in financial and foreign
exchange markets spillover into the Tanzanian economy. The analysis uses the
Cointegrated Vector Autoregressive (VAR) model as a statistical benchmark and is
generally based on data spanning the period 1980–2015.
We find a strong evidence to suggest that China’s structural rebalancing away
from commodity-intensive investment-led economy and its waning economic
growth are associated with a significant contraction in Tanzania’s exports. The
empirical estimates indicate that a 1 percentage point (ppts) lower investment
growth in China is linked with 0.6 ppts decline in Tanzania’s export growth. In
addition, long-run analysis revealed that a 1 percent contraction in China’s
domestic investment would lead to a drop in Tanzania’s exports of about 0.57
percent. These findings do not come as much of a surprise considering that China
is now the country’s third major export destination, with total Sino-Tanzania trade
surging to around $2.6 billion in 2014 from negligible levels in the early 2000s.
The rapidly increasing importance of China’s development finance to Tanzania
indicates that a slowing Chinese economy might put further strain on the domestic
economy via lower development loans and, to a limited extent, aid. However,
lower Chinese FDI is unlikely to trigger sweeping repercussions on the economy
since it accounts for just less than 1 percent of the total FDI stock in Tanzania,
among other factors.
It is, however, worth noting that the magnitude of the spillover from China’s
slowing economic growth will depend on how successful and smooth China’s
economic rebalancing will be in the years ahead. Our results generally imply that a
hard landing of the Chinese economy to its ‘new normal’ would doubtless send
shockwaves through the Tanzanian economy by driving down demand and prices
for its export commodities, and possibly lowering development finance. To shield
itself from sharp swings associated with the generalized slowdown across most

www.economics-ejournal.org 28
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

emerging market economies, the country may need to make concerted endeavors to
diversify export destinations.
In addition, we found that a 1 percent fall in export prices is associated with a
0.62 percent decrease in exports value. This attests that vulnerability to the vagaries
of world market prices remains to be the Achilles' heel of the Tanzanian economy.
Therefore, efforts targeted at diversifying the export portfolio and improving the
quality of existing products can help the country mitigate headwinds due to price
fluctuations and strengthen its external competitiveness.
The empirical results also suggest that a 1 percent increase in net private capital
flows to Tanzania contributes about 0.04 percent to national income while capital
flow volatility barely reduces growth by 0.01 ppts. The very modest impact of
increased volatility in capital flows is not too surprising given Tanzania has
relatively low level of financial development and shallow integration into the
global economy. Tanzania has so far opted for FDI over short-term portfolio equity
and debt flows, which partly explains why the country has generally remained
unscathed by previous global financial turbulences. However, as the economy is
moving towards deeper financial integration, with rising private capital flows and
external commercial borrowing as well as pending sovereign bond issuance, a
significant rise in financial market volatility may pose substantial risks in the
future.
Finally, faster nominal currency depreciation is associated with higher inflation
in Tanzania. Specifically, a 1 percent depreciation of the nominal effective
exchange rate leads to 0.58 ppts higher inflation rate. In addition, the impact of a 1
ppts drop in world oil and food price inflation rates would be to reduce overall
domestic inflation by 0.2 ppts and 0.5 ppts, respectively. Therefore, despite the
considerable depreciation of the Tanzanian Shilling since the early 2015, the
inflation rate has remained low and stable mainly due to the countervailing effects
of extraordinarily low import prices, notably that of oil. The main findings of this
paper have been shown to be sufficiently robust to a battery of sensitivity checks.

Acknowledgement The usual disclaimer applies. The views expressed in this paper are
solely those of the author and do not in any way represent the Word Bank Group.

www.economics-ejournal.org 29
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

References
Bai J., R.L. Lumsdaine, and J.H. Stock (1998). Testing for and Dating Common Breaks in
Multivariate Time Series. Review of Economic Studies 65(3): 395–432.
https://academic.oup.com/restud/article-abstract/65/3/395/1565347/Testing-For-and-
Dating-Common-Breaks-in

Doornik, J.A., and D.F. Hendry (2001). Empirical Econometric Modelling Using PcGive,
Volumes I, II and III. London: Timberlake Consultants Press.

Doornik, J.A., D.F. Hendry, and F. Pretis (2013). Step-Indicator Saturation. Discussion
Paper 658, Department of Economics, University of Oxford.
https://www.economics.ox.ac.uk/Department-of-Economics-Discussion-Paper-
Series/step-indicator-saturation

Drummond, P., and E.X. Liu (2013). Africa’s Rising Exposure to China: How Large are
Spillovers through Trade? IMF Working Paper No. 250. IMF, Washington, D.C.
https://www.imf.org/external/pubs/cat/longres.aspx?sk=41144.0

Gonzalo, J. (1994). Five Alternative Methods of Estimating Long-run Equilibrium

Relationships. Journal of Econometrics 60(1-2): 203–233.
http://www.sciencedirect.com/science/article/pii/0304407694900442

Hausmann, R., and E. Fernandez-Arias (2000). Foreign Direct Investment: Good

Cholesterol? RES Working Papers No. 417. Washington, D.C.: Inter-American
Development Bank.
https://pdfs.semanticscholar.org/c911/70e74fba30c38f19bf977adb1a40165f896b.pdf

Hausmann, R., L. Pritchett, and D. Rodrik (2005). Growth Accelerations. Journal of

Economic Growth 10(4): 303–329.
http://link.springer.com/article/10.1007/s10887-005-4712-0

Hoover, K., K. Juselius, and S. Johansen (2008). Allowing the Data to Speak Freely: The
Macroeconometrics of the Cointegrated Vector Autoregression. American Economic
Review 98(2): 251–255.
https://ideas.repec.org/a/aea/aecrev/v98y2008i2p251-55.html

Hungnes, H. (2005). Identifying the Deterministic Components in GRaM for Ox

Professional: User Manual and Documentation. Mimeo.
http://www.hungnes.net/GRaM

Hungnes, H. (2010). Identifying Structural Breaks in Cointegrated Vector Autoregressive

Models. Oxford Bulletin of Economics and Statistics 72(4): 551–565.
http://onlinelibrary.wiley.com/doi/10.1111/j.1468-0084.2010.00586.x/abstract

IMF (2012). China: Chapter IV Spillover Report, July 2012. Washington, D.C.: IMF.

IMF (2015). Sub-Sharan Africa: Dealing with the Gathering Clouds. IMF: Washington,
D.C.: International Monetary Fund, Oct. 2015.

www.economics-ejournal.org 30
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Johansen, S. (2000). Modelling of Cointegration in the Vector Autoregressive Model.

Economic Modelling 17(3): 359–373.
http://www.sciencedirect.com/science/article/pii/S0264999399000437

Johansen, S., and K. Juselius (1994). Identification of the Long-run and Short-run
Structure: An Application to the ISLM model. Journal of Econometrics 63(1): 7–36.
http://www.sciencedirect.com/science/article/pii/0304407693015595

Johansen, S., R. Mosconi, and B. Nielsen (2000). Cointegration analysis in the presence of
structural breaks in the deterministic trend. Econometrics Journal 3(2): 216–249.
http://onlinelibrary.wiley.com/doi/10.1111/1368-423X.00047/abstract

Jones, B.F., and B.A. Olken (2008). The Anatomy of Start-Stop Growth. Review of
Economics and Statistics 90(3): 582–587.
https://ideas.repec.org/a/tpr/restat/v90y2008i3p582-587.html

Juselius, K. (2006). The Cointegrated VAR model: Methodology and Applications. Oxford:
Oxford University Press.

Lakatos, C., M. Maliszews, I. Osorio-Rodarte, and D. Go (2016). China’s Slowdown and

Rebalancing: Potential Growth and Poverty Impacts on Sub-Saharan Africa. Policy
Research Working Paper 7666. World Bank, Washington, D. C.

Rahbek, A., E. Hansen, and J.G. Dennis (2002). ARCH Innovations and their Impact on
Cointegration Rank Testing. Working Paper no. 22, Centre for Analytical Finance,
University of Copenhagen.

Reinert, K.A., R.S. Rajan, A.J. Glass, and L.S. Davis (2010). The Princeton Encyclopedia
of the World Economy. P394. New Jersey: Princeton University Press.

Spanos, A. (2009). The Pre-eminence of Theory versus the European CVAR Perspective in
Macroeconometric Modeling. Economics: The Open-Access Open Assessment E-
Journal 3(2009-10).
http://www.economics-ejournal.org/economics/journalarticles/2009-10

Wacziarg, R., and K.H. Welch (2008). Trade Liberalization and Growth: New Evidence.
World Bank Economic Review 22(2): 187–231.
https://academic.oup.com/wber/article/22/2/187/1736899/Trade-Liberalization-and-
Growth-New-Evidence

World Bank (2015a). Global Economic Prospects: The Global Economy in Transition.
World Bank: Washington, D.C.

World Bank (2015b). Why Should Tanzanians Pay Taxes? The Unavoidable Need to
Finance Economic Development. Seventh Tanzania Economic Update. Dar es
Salaam, Tanzania: World Bank.

www.economics-ejournal.org 31
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

World Bank (2016). The Road Less Traveled: Unleashing Public Private Partnerships in
Tanzania. Eighth Tanzania Economic Update. Dar es Salaam, Tanzania: World
Bank.

Appendix

Appendix Table 1. Impact of commodity prices on FDI inflows (1990–2014)

(Accepted with a p-value of 0.94)
Long-run Error correction
relations coefficients
Var. 𝛽̂1 𝛽̂2 𝛽̂3 𝛼�1 𝛼�2 𝛼�3
ex𝑡 1.00 − − −0.59 −1.58 −0.12
(−5.77) (−7.18) (−4.95)
cdi𝑡 −0.43 −0.60 − ∗ ∗ ∗
(−5.53) (−11.29)
price𝑡 −0.76 1.00 −3.22
∗ ∗ ∗
(−5.61) (−12.57)
yworld − −1.65 −
𝑡 ∗ ∗ ∗
(−6.02)
fdi𝑡 − − 1.00 ∗ −5.51 −0.76
(−2.89) (−3.61)
t − 0.08 −
(3.65)
t2002 − 0.08 0.62
(3.65) (6.26)

Note: t-values in parentheses. *Denote statistically insignificant adjustment coefficients (𝑡-value less than 1.80).

www.economics-ejournal.org 32
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Appendix Table 2. Impact of China’s slowdown (1990–2014)

Cointegration rank 𝑟 = 3 No impulse dummy
(Accepted with a p-value of 0.94) (Accepted with a p-value of 0.79)
Long-run Error correction Long-run Error correction
relations coefficients relations coefficients
Var. 𝛽̂1 𝛽̂2 𝛼�1 𝛼�2 𝛽̂1 𝛽̂2 𝛼�1 𝛼�2
ex𝑡 1.00 − −0.77 −0.75 1.00 − −0.72 −0.65
(−4.79) (−4.54) (−4.79) (−4.54)
cdi𝑡 −0.54 −0.97 ∗ ∗ −0.65 −1.17 ∗ ∗
(−5.46) (−3.59) (−12.50) (−3.90)
price𝑡 −0.60 1.00 ∗ −0.49 −0.44 1.00 ∗ -0.50
(−3.35) (−1.98) (−3.21) (-2.19)
yworld
𝑡 − −3.13 ∗ ∗ − −7.96 ∗ ∗
(−2.55) (−4.22)
t − 0.08 − 0.36
(3.65) (4.51)
t2002 − −0.11 − −0.12
(2.55) (6.67)

Note: t-values in parentheses. *Denote statistically insignificant adjustment coefficients (𝑡-value less than 1.80).

Appendix Table 3. Impact of China’s slowdown (1990−2014)

Using GDP as a proxy for domestic developments in Modelling world income as an endogenous
China variable
(Accepted with a p-value of 0.45) (Accepted with a p-value of 0.97)
Long-run Error correction Long-run Error correction
relations coefficients relations Coefficients
Var. ̂
𝛽1 ̂
𝛽2 𝛼� 1 𝛼�2 ̂
𝛽1 ̂
𝛽2 𝛼
�1 𝛼�2
𝑒𝑒𝑡 1.00 − −0.67 −0.48 1.00 − −0.77 −0.74
(−2.89) (−2.52) (−4.44) (−4.95)
𝐺𝐺𝐺𝑡 −0.95 −1.55 ∗ ∗ −0.67 −1.01 ∗ ∗
(−9.29) (−1.99) (−12.27 (−3.87)
)
𝑝𝑝𝑝𝑝𝑝𝑡 −0.38 1.00 ∗ −0.61 −0.54 1.00 ∗ −0.44
(−2.07) (−2.71) (−4.51) (−2.09)
𝑦𝑡𝑤𝑤𝑤𝑤𝑤 − −5.01 ∗ ∗ − −7.46 ∗ ∗
(−4.72) (−5.48)
𝑇 − 0.28 − 0.32
(2.96) (5.05)
𝑡2002 − −0.11 − −0.12
(−5.30) (−7.08)

Note: t-values in parentheses. *Denote statistically insignificant adjustment coefficients (𝑡-value less than
1.80).

www.economics-ejournal.org 33
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Appendix Figure 1. The variables of Model 1 in levels (logs)

Appendix Figure 2. The variables of Model 1 in first differences

www.economics-ejournal.org 34
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Appendix Figure 3: The variables of Model 2 in levels (logs)

Appendix Figure 4. The variables of Model 2 in first differences

www.economics-ejournal.org 35
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Appendix Figure 5: The variabl of Model 3 in levels logs

Appendix Figure 6. The variables of Model 3 in first differences

www.economics-ejournal.org 36
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Appendix Figure 7. Export price (in level and first difference)

Structural breaks determined by Step-Indicator Saturation (SIS) and Impulse-Indicator Saturation
(IIS) (with breaks selected at 𝛼 = 0.01)

Appendix Figure 8. GDP (in level and first difference)

Structural breaks determined by SIS and IIS (with breaks selected at 𝛼 = 0.01)

www.economics-ejournal.org 37
 Economics: The Open-Access, Open-Assessment E-Journal 11 (2017–9)

Appendix Figure 9. Inflation (in level and first difference)

Structural breaks determined by SIS and IIS (with breaks selected at 𝛼 = 0.01)

Appendix Figure 10. Money supply (in level and first difference)
Structural breaks determined by SIS and IIS (with breaks selected at 𝛼 = 0.01)

www.economics-ejournal.org 38
Please note:
You are most sincerely encouraged to participate in the open assessment of this article.
You can do so by posting comments.
Please go to:
http://dx.doi.org/10.5018/economics-ejournal.ja.2017-9

The Editor

© Author(s) 2017. Licensed under the Creative Commons License - Attribution 4.0 International (CC BY 4.0).