For the statistical analysis we used the numpy, scipy, networkx and matplotlib libraries of Python 3. In specific, we used the numpy polyfit method and the stats methods of scipy for the MLE, the KS test of goodness of fit and the calculation of the Spearman correlation coefficients (for details see the methods section). We used the networkx library for the computation of centrality measures, including degree, betweeness, closeness, and the authority and hub scores. Finally, we used the matplotlib library for visualizatons including the scatterplots. Following the input-output model \cite{Leontief1936}, production is computed as: \begin{equation} x = (I-A)^{-1}f = Lf \end{equation} where $x$ is the $n \times 1$ column vector of output, $A= [a_{ij}]$ is the $nxn$ matrix of direct input coefficients, $f$ is the $nx1$ column vector of final demand, and $L=(I-A)^{-1}$ is an $n \times n$ matrix known as the Leontief inverse or the total requirements matrix. We compute aggregate effect as in the impact analysis in the input-output literature. Aggregate effect is how much production is needed to compensate a change in final demand of sector $i$. This is also known as the output multiplier, which is computed as the ith column sum of the $L$ matrix, and represent a sector-to-economy multiplier that relates final demand in sector $i$ to economy-wide output \cite{MillerBlair2009}. To compute aggregate effect we calculate production before and after the shock. The effect of a shock is defined as the change in output that was needed to compensate the shock, this is: \begin{equation} \Delta x=L \Delta f \end{equation} How many other sectors, and by how much, will be affected? If only one or a few sectors are affected, then the effect is concentrated. But, on the contrary, if the effect is spread among several sectors with a skewed distribution of the magnitudes of the effects, then the effect is heterogeneously diffused. We focus on the individual effects in each case: supposed sector i experienced a shock on final demand, this will have an effect on production on the other sectors connected directly or indirectly to $i$: $j=1,2,...,n$. The aggregate effect is the output multiplier and is computed as the sum of the $\Delta x_{j}$. This aggregate effect can be disaggregated to observe the chain of effects that took place; for that we disaggregate $\Delta x_{j}=L \Delta f$ as: \begin{equation} \Delta x_{j} = x^{1}_{j} - x^{0}_{j} = \sum^{n}_{k=1} l_{jk}f^{0}_{k} \end{equation} \begin{equation} \Delta x_{j} = (l_{j1}f^{1}_{1} + l_{j2}f^{1}_{2} + \hdots + l_{jn}f^{1}_{n})-(l_{j1}f^{0}_{1} + l_{j2}f^{0}_{2} + \hdots + l_{jn}f^{0}_{n}) \end{equation} \begin{equation} \Delta x_{j} = l_{j1} \Delta f_{1} + l_{j2} \Delta f_{2} + ... + l_{jn} \Delta f_{n} \end{equation} We take each $\Delta x_{j}$ its see its distribution. This gives the disaggregation of the effect on sector $j$'s production when sector i received a shock. According to $j$, the distribution will be different. We perform this analysis for all sectors in the economy. This will give information on how the effect of a shock is distributed along the production process the shocked sector is embedded in. To compute a measure of the diffusion of the effect of a shock we use the Leontief inverse. The diffusion measure is based on the Herfindahl concentration index \cite{Herfindahl1950}. This index for the input-output network is defined as follows \cite{AlatristeContreras2014}: \begin{equation} h_{i} = \sum^{n}_{i=1}s^{2}_{i} = \sum^{n}_{j=1}(\frac{l_{ji}}{\sum^{n}_{j=i}l_{ji}})^{2} \end{equation} where $s_{i}$ is the fraction of the effect of the shock absorbed by sector $i$ in each case, $l_{ik}$ is the $jk-th$ element of the Leontief inverse, and n is the number of sectors in the economy. To capture the diffusion properties of sectors, we computed $(1-H_{i})$ for each sector, where a high $(1-H_{i})$ means good diffusion properties or low concentration of the effect. In contrast to the aggregate effect and the backward and forward linkages found in the input-output literature, the diffusion measure we propose disaggregates the effect and provides the different distributions of the effect of the shock according to the sector that was originally shocked. We then rank the sectors according to production, exports, value added, aggregate effect and diffusion. We identify not only which sectors generated the highest aggregate effect and widest diffusion, but we study the chain of effects and the different sectors that were affected in each case. To associate the strategic sectors according to the macroeconomic variables and the diffusion to the structural properties of the economic system, we represent the economy as a network and compute global centralities. A network is a graph $G(V,E)$, composed of a set nodes or vertices $v_{i} \in V$ connected by a set of edges or links $e_{ij}\in E$. The direct input coefficients matrix, $A$, gives rise to a weighted directed graph with self-loops. The $A$ matrix is the weighted adjacency matrix of the network. Every row of the input-output table is a node in the network and every cell represents a weighted link between nodes. Values on the diagonal of the $A$ matrix represent self-loops and capture a sector using its own output as input. We compute global centralities that rank the sectors considering the different direct and indirect linkages appropriately weighted according to their importance. The centrality measure we use generates an authority and a hub score for each sector corresponding to the directionality of the link. These scores are a generalization of the eigenvector centrality for directed networks.\footnote{Eigenvector centrality was first proposed in \cite{Bonacich1987} as a power measure in social networks.} The eigenvector centrality gives each node a centrality that depends both on the number and the quality of its connections, where a node with a smaller number of high-quality links may outrank one with a larger number of mediocre links \cite{Newman2010}. Authority and hub scores were originally applied to study the link structure of web pages through the HITS (Hypertext Induced Topic Selection) algorithm.\footnote{In this context, a good hub is a page that redirects to pages that are the most prominent sources of primary information (good authorities) and hubs assemble high-quality guides and resource lists directing the users of web pages to recommended authorities \cite{Kleinberg1999}.} This algorithm assigns a hub score $h_{i}$ and an authority score $a _{i}$ to each node or web page. We thus have a circular definition of hubs and authorities. This is turned into an iterative computation, where initially $h(v)=a(v)=1$. The core of the process is a pair of updates to the hub and authority scores of all nodes \cite{Kleinberg1999}: $\mathbf{\vec{h} \leftarrow A\vec{a}}$ and $\mathbf{\vec{a} \leftarrow A^{T} \vec{h}}$. Substituting we obtain: $\mathbf{\vec{h} \leftarrow AA^{T}\vec{a}}$ and $\mathbf{\vec{a} \leftarrow A^{T}A\vec{h}}$. Iterating the process, the sequence of vectors converges to a limit or fixed point were $a^{*}$ is the principal eigenvector of $\mathbf{A^{T}A}$, and $h^{*}$ is the principal eigenvector of $\mathbf{AA^{T}}$. The the scores are given by the following equations: \begin{equation} \mathbf{a = (I- \lambda A^{T}A)^{-1}1} \end{equation} \begin{equation} \mathbf{h = (I- \lambda AA^{T})^{-1}1} \end{equation} A high hub score means that the sector is a good supplier of intermediate goods and a high authority score means a sector ir a good buyer of intermediate goods. the scores we obtain are used to rank the sectors, where the highest rank is for the most central sector. To investigate the association between variables we calculated the Spearman correlation coefficient between pair of variables. This, together with scatterplots, help elucidating the relations between macroeconomic variables and properties of sectors. For certain relations, we also fitted the data to a polynomial to better understand the relation. For the fitting process we used the polyfit function of numpy (Python) which uses a polynomial to approximate the curve: $p(x_{j})= y_{j}$, where $p(x_{j})$ is the polynomial and $y_{j} $is the data. The function minimizes the squared error: $E= \sum (|p(x_{j}) - y_{j}|)^{2}$.