Michael Cameron (p), Mohana Mondal, Jacques Poot

Projecting populations for small areas or small sub-groups of the population presents a particular challenge. The cohort-component model is insufficient for these purposes because of the lack of sufficient data on the past components of population change (births, deaths, migration). The complexity and difficulty of the projection task is compounded when projections of both small ethnic groups and small geographic areas are combined. However, an understanding of the future spatial diversity of neighbourhoods is desirable from both policy and planning standpoints.

New Zealand is incredibly ethnically diverse. In the 2013 Census, Statistics New Zealand recorded over 80 ethnic groups that each had at least 1,000 members, in a total population of around 4.2 million. Auckland, New Zealand’s largest city, has a population of 1.4 million, and is considered to be a super-diverse city.

In this paper, we report on the development of a prototype spatial microsimulation model for projecting small area ethnic populations. The model operates at the ‘area unit’ level (approximately suburbs), and considers all 21 ethnic groups at Level 2 of New Zealand’s standard classification of ethnicities.

The basic structure of the model includes modules that estimate: (1) locational transition probabilities (migration); and (2) inter-ethnic transition probabilities (inter-ethnic mobility). We report preliminary results from the prototype model, and derive some implications for future development of the model, as well as more general learnings that can be applied to the development of other similar models.

Bartlomiej Rokicki (p), Krzysztof Ostaszewski , Solomiya Yushchak

Since its EU accession, major infrastructure projects have been one of the main regional policy tools in Poland. As a result, between 2004 and 2015 there were 240 projects with budget over euro 50 million each and over 130 projects with budget over euro 100 million each. Yet, the experience of other countries shows that the implementation of big projects usually leads to many problems and ends up with ineffective spending of public funds. In particular, Flyvbjerg (2003, 2014) argues that the big infrastructure project delivery is highly problematic with a dismal performance record in terms of actual costs and benefits.
This paper applies actuarial credibility approach to analyze viability of major infrastructure projects in terms of the cost overruns and delays. In particular we verify whether the amount of financial means devoted to the big projects is adequate as compared to the real costs associated with their accomplishment. We compare the data on projects co-financed by EU funding with the ones fully financed from national sources. We find that the performance of major infrastructure investments depends on the source of funding, type of investment or region of implementation. We show that actuarial credibility approach allows to foresee the degree of possible overruns and delays before the project implementation.

Malte Jahn, Anthony Flegg,Timo Tohmo (p)

ABSTRACT (300 words)
The multipliers obtainable from a regional input-output (I-O) table are a valuable analytical tool, yet such tables typically must be constructed via non-survey methods. Although Flegg’s location quotient (FLQ) is a method that often performs well, it is designed to estimate intraregional intermediate transactions and coefficients. The input coefficients for different regions are estimated independently and interregional coefficients are not estimated explicitly.
A dataset constructed by the Bank of Korea for all 16 South Korean regions in 2005 is one of the few survey-based full interregional I-O tables. It has data for all intersectoral transactions, both within and across regions, thereby allowing us to test some alternative theoretical approaches. Our focus is on an innovative methodological approach proposed by Jahn (2017), in which two methods of estimation, the FLQ and a gravity model, are combined in a consistent way to estimate the intraregional and interregional transactions, respectively. All regions are treated simultaneously. Furthermore, the estimated transactions are constrained to equal the national aggregates for each pair of sectors.
A novelty of our paper is its use of statistical information criteria to determine the best model for estimating output multipliers. Such criteria are relevant when the approaches being compared employ very different numbers of parameters. With the FLQ, for instance, one has a choice between pursuing a simple approach, whereby an unknown parameter δ is held to be invariant across both sectors and regions, and more complex approaches where these assumptions are relaxed. Standard performance criteria cannot reveal whether the inclusion of extra parameters is warranted, whereas information criteria can do so. We demonstrate that, for South Korea, the best approach is to combine the FLQ with a simple trade model. Since the interregional trade flows do not seem to depend much on distances or adjacency, a gravity model is unnecessary.

Piotr Wójcik (p)

The analysis of real economic convergence and its factors is one of the important topics of research in the field of macroeconomics. It also has important implications for economic policy. The purpose of the article is to identify factors of income convergence between countries with the use of regularized LASSO regression. According to the author's best knowledge, this algorithm has never been used in growth regressions before.

Durlauf et al. (2009) indicate difficulties in verifying conditional beta convergence – the choice of control variables has a key impact on the inference about its occurrence. There is no consensus on what their set is the best. Conclusions regarding the significance of individual factors may contradict each other.

Different approaches were used to select a best subset of convergence factors. Sala-i-Martin (1997) considered all combinations of 62 variables and estimated two million regressions, measuring the significance of individual explanatory variables by weighted statistics based on all regressions. This approach was criticized by Hendry and Krolzig (2004), who indicate that to correctly identify statistically significant convergence factors, it is sufficient to estimate one regression and apply the “general to the specific” approach. Another alternative makes a use of bayesian model averaging – applied to growth regressions by Sala-i-Martin et al. (2004) who called this specific approach Bayesian Averaging of Classical Estimates.

In this article an alternative tool is used, namely the LASSO method (Tibishrani, 1996). It is a popular machine learning tool often used for pre-selection of potentially important explanatory variables. It can also be used when the initial number of variables exceeds the number of observations. In the case of modeling a continuous variable using the LASSO method, the cost function in the optimization problem apart from minimizing the sum of squared residuals also takes into account the sum of absolute values of the model parameters as an additional constraint. At the expense of a certain bias of the obtained parameter estimates, LASSO often allows to obtain more precise forecasts on the test sample and, what is important in the context of this research, select the most important factors of the studied phenomenon by eliminating the excess variables from the model. The presented article uses leave-one-out cross validation for selection of the optimal lambda hyperparameter (the weight of the additional constraint). The algorithm is applied on empirical data from Sala-i-Martin X. (1997) and Fernández, Ley and Steel (2001), among others.