Sex Hormones, Growth Hormone, and the Cornea

The growth and maintenance of nearly every tissue in the body is influenced by systemic hormones during embryonic development through puberty and into adulthood. Of the ~130 different hormones expressed in the human body, steroid hormones and peptide hormones are highly abundant in circulation and are known to regulate anabolic processes and wound healing in a tissue-dependent manner. Of interest, differential levels of sex hormones have been associated with ocular pathologies, including dry eye disease and keratoconus. In this review, we discuss key studies that have revealed a role for androgens and estrogens in the cornea with focus on ocular surface homeostasis, wound healing, and stromal thickness. We also review studies of human growth hormone and insulin growth factor-1 in influencing ocular growth and epithelial regeneration. While it is unclear if endogenous hormones contribute to differential corneal wound healing in common animal models, the abundance of evidence suggests that systemic hormone levels, as a function of age, should be considered as an experimental variable in studies of corneal health and disease.

Emerging Markets Interest Rates, International Reserves and Net Foreign Assets

In the context of a managed float regime, we adopt the portfolio balance view to show the effects of the net foreign assets of an economy and its gross international reserves level on interest rate differentials. We argue that the interest rate differential can be explained by three components, where the components are the expected depreciation of the domestic currency, a default risk premium, and a portfolio balance premium. Our theoretical analysis suggests that the interest differential is a convex function of the level of gross international reserves. In particular, the differential and gross reserves are inversely related at low levels of reserves, but positively at higher levels. We evaluate our framework for the case of Lebanon. We find that the differential is inversely related to both net foreign assets and gross international reserves. These findings are then confirmed with data from Indonesia and Mexico.

Reinforcement Learning for Systematic FX Trading

We conduct a detailed experiment on major cash fx pairs, accurately accounting for transaction and funding costs. These sources of profit and loss, including the price trends that occur in the currency markets, are made available to our recurrent reinforcement learner via a quadratic utility, which learns to target a position directly. We improve upon earlier work, by casting the problem of learning to target a risk position, in an online learning context. This online learning occurs sequentially in time, but also in the form of transfer learning. We transfer the output of radial basis function hidden processing units, whose means, covariances and overall size are determined by Gaussian mixture models, to the recurrent reinforcement learner and baseline momentum trader. Thus the intrinsic nature of the feature space is learnt and made available to the upstream models. The recurrent reinforcement learning trader achieves an annualised portfolio information ratio of 0.52 with compound return of 9.3\%, net of execution and funding cost, over a 7 year test set. This is despite forcing the model to trade at the close of the trading day 5pm EST, when trading costs are statistically the most expensive. These results are comparable with the momentum baseline trader, reflecting the low interest differential environment since the the 2008 financial crisis, and very obvious currency trends since then. The recurrent reinforcement learner does nevertheless maintain an important advantage, in that the model's weights can be adapted to reflect the different sources of profit and loss variation. This is demonstrated visually by a USDRUB trading agent, who learns to target different positions, that reflect trading in the absence or presence of cost.<br>

Reinforcement Learning for Systematic FX Trading

We conduct a detailed experiment on major cash fx pairs, accurately accounting for transaction and funding costs. These sources of profit and loss, including the price trends that occur in the currency markets, are made available to our recurrent reinforcement learner via a quadratic utility, which learns to target a position directly. We improve upon earlier work, by casting the problem of learning to target a risk position, in an online learning context. This online learning occurs sequentially in time, but also in the form of transfer learning. We transfer the output of radial basis function hidden processing units, whose means, covariances and overall size are determined by Gaussian mixture models, to the recurrent reinforcement learner and baseline momentum trader. Thus the intrinsic nature of the feature space is learnt and made available to the upstream models. The recurrent reinforcement learning trader achieves an annualised portfolio information ratio of 0.52 with compound return of 9.3\%, net of execution and funding cost, over a 7 year test set. This is despite forcing the model to trade at the close of the trading day 5pm EST, when trading costs are statistically the most expensive. These results are comparable with the momentum baseline trader, reflecting the low interest differential environment since the the 2008 financial crisis, and very obvious currency trends since then. The recurrent reinforcement learner does nevertheless maintain an important advantage, in that the model's weights can be adapted to reflect the different sources of profit and loss variation. This is demonstrated visually by a USDRUB trading agent, who learns to target different positions, that reflect trading in the absence or presence of cost.<br>

Reinforcement Learning for Systematic FX Trading

We conduct a detailed experiment on major cash fx pairs, accurately accounting for transaction and funding costs. These sources of profit and loss, including the price trends that occur in the currency markets, are made available to our recurrent reinforcement learner via a quadratic utility, which learns to target a position directly. We improve upon earlier work, by casting the problem of learning to target a risk position, in an online learning context. This online learning occurs sequentially in time, but also in the form of transfer learning. We transfer the output of radial basis function hidden processing units, whose means, covariances and overall size are determined by Gaussian mixture models, to the recurrent reinforcement learner and baseline momentum trader. Thus the intrinsic nature of the feature space is learnt and made available to the upstream models. The recurrent reinforcement learning trader achieves an annualised portfolio information ratio of 0.52 with compound return of 9.3\%, net of execution and funding cost, over a 7 year test set. This is despite forcing the model to trade at the close of the trading day 5pm EST, when trading costs are statistically the most expensive. These results are comparable with the momentum baseline trader, reflecting the low interest differential environment since the the 2008 financial crisis, and very obvious currency trends since then. The recurrent reinforcement learner does nevertheless maintain an important advantage, in that the model's weights can be adapted to reflect the different sources of profit and loss variation. This is demonstrated visually by a USDRUB trading agent, who learns to target different positions, that reflect trading in the absence or presence of cost.<br>

Reinforcement Learning for Systematic FX Trading

We conduct a detailed experiment on major cash fx pairs, accurately accounting for transaction and funding costs. These sources of profit and loss, including the price trends that occur in the currency markets, are made available to our recurrent reinforcement learner via a quadratic utility, which learns to target a position directly. We improve upon earlier work, by casting the problem of learning to target a risk position, in an online learning context. This online learning occurs sequentially in time, but also in the form of transfer learning. We transfer the output of radial basis function hidden processing units, whose means, covariances and overall size are determined by Gaussian mixture models, to the recurrent reinforcement learner and baseline momentum trader. Thus the intrinsic nature of the feature space is learnt and made available to the upstream models. The recurrent reinforcement learning trader achieves an annualised portfolio information ratio of 0.52 with compound return of 9.3\%, net of execution and funding cost, over a 7 year test set. This is despite forcing the model to trade at the close of the trading day 5pm EST, when trading costs are statistically the most expensive. These results are comparable with the momentum baseline trader, reflecting the low interest differential environment since the the 2008 financial crisis, and very obvious currency trends since then. The recurrent reinforcement learner does nevertheless maintain an important advantage, in that the model's weights can be adapted to reflect the different sources of profit and loss variation. This is demonstrated visually by a USDRUB trading agent, who learns to target different positions, that reflect trading in the absence or presence of cost.<br>

Variance Risk Premia, Asset Predictability Puzzles, and Macroeconomic Uncertainty

This article reviews the predictability evidence on the variance risk premium: ( a) It predicts significant positive risk premia across equity, bond, currency, and credit markets; ( b) the predictability peaks at few-month horizons and dies out afterward; ( c) such a short-run predictability is complementary to the long-run predictability offered by the price-to-earnings ratio, forward rate, interest differential, and leverage ratio. Several structural approaches based on the notion of economic uncertainty are discussed for generating these stylized facts about the variance risk premium, which has broad implications for various empirical asset pricing puzzles.

Econometric testing of uncovered interest rate parity in Serbia

This paper provides econometric evidence of the interest parity puzzle in Serbia over the period 2005-2016. Econometric findings are derived from the following techniques: long-run parameter estimation based on the autoregressive distributed lag model, impulse response function computed from the bivariate vector autoregressive model, and estimation of the two-regime Markov switching parameter model. Our results indicate that a positive interest differential corrected for country risk leads to significant dinar appreciation against the euro. The intensity of this impact is different across sub-periods of low exchange rate variability and high variability. Exchange rate movements are found to appreciate more strongly during lower variability episodes. Preliminary econometric investigation of four other European emerging economies documents similar findings only for Romania. Our results suggest that there is a huge incentive for shortterm carry trades in Serbia, regardless of substantial risks.

FUSION OF LOCATION FINGERPRINTING AND TRILATERATION BASED ON THE EXAMPLE OF DIFFERENTIAL WI-FI POSITIONING

Positioning of mobile users in indoor environments with Wireless Fidelity (Wi-Fi) has become very popular whereby location fingerprinting and trilateration are the most commonly employed methods. In both the received signal strength (RSS) of the surrounding access points (APs) are scanned and used to estimate the user’s position. Within the scope of this study the advantageous qualities of both methods are identified and selected to benefit their combination. By a fusion of these technologies a higher performance for Wi-Fi positioning is achievable. For that purpose, a novel approach based on the well-known Differential GPS (DGPS) principle of operation is developed and applied. This approach for user localization and tracking is termed Differential Wi-Fi (DWi-Fi) by analogy with DGPS. From reference stations deployed in the area of interest differential measurement corrections are derived and applied at the mobile user side. Hence, range or coordinate corrections can be estimated from a network of reference station observations as it is done in common CORS GNSS networks. A low-cost realization with Raspberry Pi units is employed for these reference stations. These units serve at the same time as APs broadcasting Wi-Fi signals as well as reference stations scanning the receivable Wi-Fi signals of the surrounding APs. As the RSS measurements are carried out continuously at the reference stations dynamically changing maps of RSS distributions, so-called radio maps, are derived. Similar as in location fingerprinting this radio maps represent the RSS fingerprints at certain locations. From the areal modelling of the correction parameters in combination with the dynamically updated radio maps the location of the user can be estimated in real-time. The novel approach is presented and its performance demonstrated in this paper.